Voice coding on a Mac has two halves: dictating your prompt, and hearing the agent's answer. For dictation you can use Claude Code's own /voice, macOS Dictation, or an app like FluidVoice. For listening back there's a DIY Stop hook, open-source tools like Aloud, flowcode and VoiceMode, or an app like Pintail. Most people set up the first half and forget the second.
I dictate my prompts and listen to the replies every day. Here's the landscape as I see it, with every tool checked against its own page on 30 September 2026.
Half one: dictate in
This half is well covered. Pick what fits how you work.
- Claude Code's
/voice. Built in. Hold or tap a key and your speech is transcribed into the prompt. It needs a Claude.ai login, and the audio goes to Anthropic's servers for transcription. It doesn't use your tokens. (docs) - macOS Dictation. Built into every Mac and works in any app, the terminal included. (Apple's guide)
- FluidVoice. What I use. Free, open source, runs on the Mac, installs with Homebrew, macOS 15 or later. It cleans up rambling speech into a usable prompt. (altic.dev/fluid)
Dictation is where most "voice coding" articles stop. But dictating a prompt and then reading a 40-line reply isn't a voice workflow. It's half of one.
Half two: listen back
Claude Code has no built-in way to read replies aloud. These are the options:
| Tool | Voice | Short summaries | Price |
|---|---|---|---|
| DIY Stop hook | System voice (say) |
No | Free |
| Aloud | Local (Kokoro) | Yes, a short gist | Free, MIT |
| flowcode | Local (Kokoro) | No, reads the prose | Free, MIT, beta |
| VoiceMode | Local or cloud | Two-way conversation | Free, MIT |
| SpeakIt | See its README | See its README | Free |
| Pintail | Cloud, on your OpenRouter account | Yes, four lengths | €15 once (planned) + ½–2 ¢ a reply |
Checked on 30 September 2026. The full comparison has more columns and when to pick each.
A few honest one-liners:
- The DIY hook is 20 lines and teaches you how Claude Code's hooks work. It reads markdown literally and tabs talk over each other.
- Aloud is the one I'd point to if you want free and fully local. It also speaks up when an agent asks a question or needs approval.
- flowcode does both halves in one local app, and speaks English and Czech.
- VoiceMode is a conversation, not a reader. I tried it and removed it because I already had dictation I liked.
- Pintail is mine. It's built for several sessions at once: short summaries, one voice at a time, and a pill that shows which project is talking.
Why the second half matters more than it looks
With one agent, reading the reply is fine. With five or six tabs, it's the bottleneck. You dictate a prompt in one tab, switch to another, and a minute later something finishes somewhere. Which tab? Did it work? Does it need you?
Hearing "review-app: the empty state is fixed and all 14 tests pass" answers all three without looking. For me, that was the missing half.
What a full voice setup looks like
Mine, for what it's worth:
- FluidVoice for dictation.
- Claude Code in iTerm2, one tab per project.
- Pintail reading the replies as short summaries, with the project's name.
The written reply is always still there. Voice doesn't replace reading the diff. It means I read it when I choose to, not because it's the only way to find out what happened.
What does it cost?
Dictation can be free (all three above). Listening back is free with the open-source tools. With a cloud voice it's about half a cent to 2 cents per reply; I wrote up the measured prices.