AI NewsDev toolsAnnouncement

gpt-voice turns your ChatGPT sign-in into Mac dictation, no API key

A small open-source menu-bar app for macOS. It sends what you say to ChatGPT and pastes the text back in your editor.

AI News

Editorial2 min read

LinkedInX
Signal field artwork for a Mac menu-bar dictation tool.

Why it mattersMany Mac users already pay for ChatGPT. This tool lets them use that sign-in for dictation. No second bill.

Many Mac users already pay for ChatGPT and would like dictation that uses that sign-in. gpt-voice is a small menu-bar app that does exactly that.

The entropyconquers team at GitHub published gpt-voice under MIT on 20 October 2026. The GitHub API reports 83 stars on 23 October 2026. The app sits in the menu bar on macOS 13 or later. The user presses a keyboard shortcut, speaks, and the app pastes the text into whatever window has focus. The default shortcut is Control, Option and Space.

How it signs in

The README is clear about one thing. The app does not use the OpenAI API. It uses the sign-in file that T3 Code, the former Codex CLI, already keeps on the user's machine. Anyone who runs Codex CLI with a ChatGPT Plus, Pro or Team account can use gpt-voice. No new key is needed. No new bill is added.

The app sends the audio to a private ChatGPT endpoint for speech to text. The README says this is not an offline tool. It needs the internet. It also says the endpoint can change at any time. The author states this plainly. The app is not affiliated with OpenAI. The sign-in is used in a way the vendor has not published a contract for. A reader has to be ready for the trick to stop working.

What it does on the Mac

The app lives in the menu bar. A click shows a small window. The user can change the shortcut, pick a different microphone and change the model used. The audio is sent only while the shortcut is held down. When the shortcut is released the text is pasted into the active app. It works in any text field: a code editor, a browser, a terminal, a note.

The app is 83 stars old. The repository has eight commits. There are no releases yet. The main branch was last pushed on 20 October 2026. There is no automated test suite in the repository. The license is MIT.

Why this matters today

Native dictation on macOS is slow and often wrong. The paid tools are a second subscription. This project is a small, honest build that borrows what the user already pays for.

The pattern is useful for any team releasing a desktop tool next to a cloud service. The hard part is the first sign-in. Letting a desktop app read the token that another CLI already stored is a short way around it. It is also fragile, for the same reason: the shape of that token file is not a public contract. The author handled this by naming T3 Code as the dependency and marking the risk in the README.

For a Reveneau reader, the item is a reminder. A small tool with 83 stars at three days old can still earn its place on a Mac this week. The cost to try it is one make command and a working ChatGPT sign-in. The cost of being wrong is small too. If the private endpoint goes away, the app stops working and nothing else breaks.

Source

entropyconquers/gpt-voice on GitHub, with stars and dates confirmed against the GitHub API.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project