maxrubin629/codex-voice

Codex Voice is a local desktop application that lets you control Codex (an AI coding assistant) by speaking to it instead of typing. It connects two things: OpenAI's Realtime voice system, which handles speech recognition and the back-and-forth audio conversation, and Codex's app-server mode, which handles actually running code, managing files, asking approval questions, and doing the coding work on your machine. The app shows a compact window where you speak your request. From there, it handles the full session lifecycle: creating and resuming projects, summarizing what Codex has done, interrupting or redirecting it mid-task, and forwarding any approval questions, like whether it is okay to delete a file, back to you through the voice interface. Each project gets its own workspace folder stored locally. You would use this if you want a hands-free or voice-first way to work with Codex on coding tasks, for example, when your hands are busy or you prefer talking through a problem rather than typing it out. It requires Node.js, the Codex command-line tool, and an OpenAI API key. The project is written in TypeScript and uses a React-based desktop UI with separate processes for the main application logic and the browser-side voice client.