Claude Code's /voice or a system-wide dictation app?
Use /voice if the Claude Code prompt is the only place you type, you sign in with a Claude.ai account, and sending audio to Anthropic is fine. Use a system-wide dictation app if you also type in the shell, an editor, a browser or Slack, if you use an API key or Bedrock, if your audio must stay on the Mac, or if you want to answer Claude's questions by voice.
What /voice is
Claude Code's voice dictation, documented at code.claude.com, is push-to-talk inside the CLI's prompt box. Run /voice, hold Space (or /voice tap for tap-to-start, tap-to-send), speak, and the transcript streams into the prompt. It needs Claude Code v2.1.69 or later, a local microphone, and a Claude.ai sign-in. Transcription is tuned for coding terms and adds your project and branch names as hints. It does not count against your usage.
The four boundaries
- Scope. It fills the prompt box and nothing else. The shell command you run to check Claude's work, the commit message, the Slack update, the Cursor prompt — all typed.
- Where the audio goes. Anthropic's documentation is plain: audio is streamed to Anthropic's servers and is not processed locally.
- Who can use it. Only with a Claude.ai account. Teams on an API key, Amazon Bedrock, Google Vertex or Microsoft Foundry do not get
/voiceat all. - Where it runs. A local microphone is required, so no SSH sessions, no cloud sessions, and in VS Code no Remote windows.
There is a fifth boundary that is not in the docs because it is not a feature: when Claude Code stops to ask you something — pick an approach, approve this plan, allow this command — /voice has no part in it. You click or type in the terminal.
What a system-wide app changes
A system-wide dictation app listens for a key anywhere on the Mac and pastes into the field in front. The prompt box becomes one of many places it works. Several Mac apps do this well, and they differ on the axes that matter for coding:
| App | Audio processed | Price | Answers Claude's cards by voice | Notes |
|---|---|---|---|---|
| FreeFlow Dictation | On your Mac | $99 once, 3 Macs | Yes (permissions click-only) | Dictionary, modes, Apple Intelligence formatting on macOS 26 |
| superwhisper | Local or cloud, by mode | $8.49/mo, $249.99 lifetime | No | Deepest mode system; FreeFlow imports its mode files |
| Spokenly | Local or cloud, by model | Free tier, paid plans | No, but an MCP integration lets the agent speak questions | Parakeet V3 free locally |
| Voibe | On your Mac | $9.90/mo, $99/yr, $199 lifetime per seat | No | Mac and Windows; developer mode for editor terminals |
| Wispr Flow | Their servers | $12–15/mo | No | Most polished; needs a connection for every sentence |
| Willow Voice | Their servers | $12–15/mo | No | Fast; generous free tier |
Choosing
- Prompt-only, Claude.ai account, audio can leave the Mac:
/voiceis free and good. Stop here. - You type in more than the prompt: any system-wide app beats
/voice, and the question becomes local or cloud. - Audio must stay on the Mac: FreeFlow, superwhisper's local modes, Spokenly's local models, or Voibe.
- API key, Bedrock, Vertex or Foundry:
/voiceis not available; a system-wide app is the only voice option. - You want to answer Claude's plans, questions and permissions out loud: FreeFlow is the one that does it, through Claude Code's hooks.
Questions people ask
Is Claude Code's /voice on-device?
No. Anthropic's documentation says audio is streamed to its servers for transcription and is not processed locally.
Does /voice work with an API key?
No. It requires a Claude.ai sign-in and is unavailable when Claude Code uses an API key, Bedrock, Vertex or Foundry.
Which languages does /voice support?
Twenty, set one at a time in /config: Czech, Danish, Dutch, English, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Turkish, Ukrainian.
Can any app answer Claude Code's permission prompt by voice?
FreeFlow shows the prompt as a card but keeps permission cards click-only on purpose. Plans and questions can be answered by voice.