Talk to Claude Code, hands-free

What Claude Code's built-in voice does

Claude Code has voice dictation: run /voice, then hold or tap Space and speak. Your words are transcribed into the prompt, tuned for coding terms. It is great at your desk. It needs a claude.ai account (not an API key), a local microphone, and you still read the answers on screen.

What OpenMimir adds

OpenMimir is a single conversation that sits in front of all your Claude Code sessions (and Codex and OpenCode, if you use them). You talk to it, it talks back, and it drives the sessions for you.

• Full duplex. It listens while it speaks, so you can interrupt. There is no key to hold.
• Spoken results. When a session finishes, OpenMimir reads the reply and tells you the outcome in a sentence or two instead of making you read a transcript.
• No ids, no picking. "How's the billing refactor?" finds the right session, including ones you started yourself in the terminal last week.
• Read-back before work. Before it starts or redirects anything by voice, it repeats the plan and waits for your "yes".
• Live captions. What it heard is on screen as you speak, so a mishearing is obvious.
• Approvals stay safe. Pushes, deploys and rm -rf need a tap on the screen; voice can't approve them.

Which one to use

	Claude Code /voice	OpenMimir
Input	Dictation into the prompt	Spoken conversation
Output	Text in the terminal	Spoken summary plus captions
Hands	Hold or tap a key	Hands-free
Sessions	The one you're in	All of them, across agents
Auth	claude.ai account	Whatever your Claude Code uses; OpenAI key for voice

They work well together: dictate at the desk, and switch to OpenMimir when you step away, ride the trainer or walk.

How it works

Voice runs on OpenAI's realtime model with your own key, straight from your browser. A separate model plans what to do and calls Claude Code headless (claude -p --resume), so the conversation continues in the same session. You can open it later with claude --resume and see everything that happened.