Anthropic added /voice to Claude Code. Here's why that's fine.

On March 3, Anthropic started rolling out voice mode for Claude Code. You type /voice, hold the spacebar, and talk. Your words become prompts. No third-party app needed.

I've been building AudioBash — a voice-controlled terminal — since December 2025. Three months of my nights and weekends. When I saw the announcement, my first thought was: well, that's that.

My second thought was: actually, no. These solve different problems.

What /voice does

Anthropic's implementation is clean. Hold spacebar, talk, release. Your speech gets transcribed and submitted as a prompt to Claude. Transcription tokens don't count against your rate limit. It's included with Pro, Max, Team, and Enterprise subscriptions at no extra cost. You can mix voice and typing in the same session — paste a file path, then speak your instructions about it.

For Claude Code users who want to talk instead of type, it works. I've used it. It's good.

So why does AudioBash still exist?

Different tools for different problems

Claude Code's /voice solves one problem: voice input for Claude Code. That's it. It transcribes your speech and sends it as a prompt to Claude.

AudioBash solves a different problem: voice input for your terminal. Any terminal. Any LLM. Any command.

Here's what I mean:

Multiple terminals

My daily workflow involves at least two terminal sessions. Claude Code in one, a dev server or test runner in another. Sometimes three or four. AudioBash has a pane system — tmux-style splits where each pane is its own shell with its own context. I can talk to Claude Code in the left pane and run tests in the right pane with the same voice shortcut. /voice only works inside Claude Code's own session.

Not just Claude

I use Gemini CLI too. And sometimes plain PowerShell for quick file operations. AudioBash works with all of them. Press Alt+S, talk, and your words go to whatever terminal has focus. It doesn't care what's running in it.

/voice is tied to Claude Code. If you switch to another tool, you lose voice input.

Agent mode

This is the feature that matters most to me. When I'm at a raw shell prompt (not inside Claude Code), I don't always know the right command. I know what I want to do in English. AudioBash's agent mode takes my speech, considers what directory I'm in and what I've been doing, and figures out the right shell command. "Find all the test files that import vitest" becomes grep -rl "import.*vitest" tests/.

/voice doesn't do this because it doesn't need to — Claude Code already understands natural language. But outside of Claude Code, at a raw prompt, you need something that translates.

Provider choice

AudioBash supports multiple transcription providers: Gemini, ElevenLabs, and local Whisper for offline use. You can pick the one that works best for you. Anthropic hasn't said what powers /voice transcription, and you can't change it.

Local Whisper matters if you're working on a plane, on slow hotel wifi, or if you just don't want your voice data going to a server.

Offline

If your internet drops, /voice stops working. AudioBash with local Whisper keeps going. I've used this on flights. It's not fast (local transcription on a laptop CPU takes a few seconds), but it works.

What we built instead of competing

When I saw the /voice announcement, the question wasn't "how do I beat this." It was "how do I make sure they work well together."

AudioBash v3.0 — which shipped the same week — includes voice mode detection. When Claude Code's /voice is active in a terminal pane, AudioBash detects it by watching the terminal output for activation patterns. When it sees /voice is running:

• The mic button shows a [CC /voice] badge so you know what's happening
• AudioBash disables its own recording on that pane to prevent interference
• If you're in a multi-pane layout, voice input automatically routes to a different pane
• If you're actively recording when /voice activates, AudioBash cancels the recording cleanly instead of sending a partial transcription

Two voice systems fighting over the same microphone would be a mess. Instead, AudioBash yields when /voice is present and stays available for everything else.

Why I'm not worried

The way people write software is changing. Vibe coding — describing what you want in natural language and letting AI figure out the implementation — is how a growing number of people work. And "vibe coders" don't just use Claude Code. They use Gemini CLI, Copilot, Cursor, Codex, and a dozen other tools. They switch between them based on the task.

Voice input for one specific tool is useful. Voice input for your entire terminal environment is a different thing.

There are also practical differences. /voice is rolling out to about 5% of Claude Code users as of this writing and is English-only. AudioBash is available now, runs locally, and supports whatever your transcription provider supports.

And there's a category difference that matters: /voice is a feature inside someone else's product. AudioBash is a standalone app you control. You pick the transcription provider. You configure the shortcuts. You own the workflow. If Anthropic changes how /voice works or limits it or removes it, that's their call. AudioBash is open source and runs on your machine.

What's next

The v3.0 release also shipped the pane system I've been wanting to build since December — binary tree splits, named sessions, keyboard navigation. It's the feature that makes AudioBash feel like a real terminal instead of a toy. Combined with voice mode detection, it means you can have Claude Code with /voice in one pane and use AudioBash voice in your other panes, all without conflicts.

I'm going to keep building this. Not because I think I can out-ship Anthropic — I obviously can't — but because AudioBash solves my specific problem in a way that /voice alone doesn't. I need multiple terminals. I need agent mode at the raw prompt. I need offline transcription on flights. I need to talk to Gemini CLI and PowerShell, not just Claude.

If you use Claude Code and /voice is enough for you, that's great. Use it. But if you want voice across your whole terminal setup, with panes and provider choice and offline support, AudioBash is still here.

And it gets along with /voice just fine.