How do I use speech to text with Codex?
Jompter gives the Codex CLI free speech to text on a Mac. Hold Right ⌘, explain the change out loud, and let go: the transcript is typed into the terminal where Codex is waiting, exactly as if you had typed it. Transcription runs on Apple Silicon with Whisper, Parakeet or Nemotron, so it works offline and your voice stays on your Mac. When the problem is visual, Jompter keeps an annotated screenshot or a screen recording in the same prompt as your words. Connect Jompter over MCP and Codex reads that whole prompt itself (the text, the screenshot paths and the key frames of a recording) and updates the prompt's to-do list as it works. The Codex CLI takes its MCP servers in TOML, so Jompter's panel writes a ~/.codex/config.toml snippet rather than JSON. Dictation, prompts and the MCP hand-off are free and unlimited; it needs Apple Silicon and macOS 13 or later.
Jompter with Codex
Talk into the Codex CLI the way you would type into it; the transcript lands where the cursor is.
The Codex CLI takes its MCP servers in TOML, so the panel generates that rather than JSON.
- MCP snippet goes in
- ~/.codex/config.toml
- Skills install to
- ~/.agents/skills/
Jompter with Codex.
The questions developers ask before trying Jompter.
Not today. Codex CLI had experimental voice transcription behind the voice_transcription feature flag from version 0.105.0, and removed it in 0.118.0. Jompter fills the gap: hold Right ⌘ and your words are typed into the terminal where Codex runs, transcribed on your Mac.
Yes. Jompter types into whatever has focus, so dictation works in Terminal, iTerm2, Ghostty or the terminal inside your editor, wherever Codex is running.
Copy the TOML snippet from Jompter's MCP panel into ~/.codex/config.toml. It already contains your port and key, and uses 127.0.0.1 because the server listens on IPv4 only.
Over MCP, yes: Codex reads the prompt with its screenshot paths and recording frames. Without MCP, paste the prompt and pass the screenshot as a file.
Yes, on your Mac. Dictation, prompts, the MCP hand-off, transforms, translation and Speak are free and unlimited. The heavier features are free within limits — 20 screenshots a day, 5 screen recordings, 3 boards and one 10-minute Remote session a day — and a $99 one-time license removes every limit.