Pintail and the alternatives

You can make coding-agent replies audible with a DIY Stop hook, macOS Spoken Content, open-source tools like Aloud, Heard, Agents Aloud, claude-speak and Kobi, or Pintail. The right one depends on whether you want it free and local, or a short summary with the session named in the notch.

I make Pintail, so read this with that in mind. I’ve tried to be fair: several of these are free, and for some people they’re the better choice. I say so below.

The comparison

OptionVoiceShort summariesShows which sessionLanguagesPriceOpen sourcePlatforms
DIY Stop hookLocal, the system voice (say, espeak)No, reads the raw textNoYour system voicesFreeYour own scriptmacOS, Linux, Windows
macOS Spoken ContentLocal, the system voiceNo, reads what you selectNoMany system voicesFree, built inNomacOS
AloudLocal (Kokoro)A short gist, the full reply on requestYes, per sessionKokoro’s voicesFreeYes (MIT)macOS; Claude Code and Codex
VoiceModeLocal (Kokoro) or cloudTwo-way conversation insteadNoDepends on the voice serviceFreeYes (MIT)macOS, Linux, Windows
SpeakySpeakLocal (Kokoro), or the system voiceNo, reads the cleaned-up replyYes: one at a time, mute per sessionEnglish; others through system voicesFreeYes (MIT)macOS 15.6+, Claude Code; builds from source
claude-voiceoverLocal, the system voiceNo, reads the prose, skips codeOn or off per sessionYour system voicesFreeYes (MIT)macOS, Linux; Claude Code plugin
flowcodeLocal (Kokoro)No, reads the proseNoEnglish, CzechFreeYes (MIT), public betamacOS; also takes dictation
HeardLocal Kokoro, or managed voices on PowerYes (progress, blockers, results)YesNot listedFree engine; Power from $24/mo yearlyEngine Apache 2.0; the app is the managed buildmacOS 14+
Agents AloudmacOS say, or ElevenLabs with your keyOptional rewriter; otherwise the messageYes, says the session nameSystem voices, or ElevenLabsFree, build from sourceYes (MIT)macOS 26+
claude-speakLocal KokoroYes, one lineYes, says the project nameEnglish summaries; Kokoro has more voicesFreeYes (MIT)macOS, Linux
KobiLocal Kokoro, or ElevenLabs with your keyYes, one line, when it needs an OK or finishesYes, says the project nameNot listedFree with Kokoro; ElevenLabs optionalYes (MIT)macOS; Claude Code plugin
Heard’s /say skill (claude-code-tts)Local, the system voice (say, spd-say, System.Speech)The Stop hook reads the raw reply; /say speaks a short summaryParallel sessions queue; the summary names the contextYour system voicesFreeYes (MIT)macOS, Linux, Windows; Claude Code and Codex
spoken-replies (claude-voice)Local, the system voice (say)No; a hook asks Claude to write the whole reply for the earNot listedYour system voicesFreeYes (MIT)macOS; Claude Code plugin
claude-read-aloudSystem voices or Kokoro; Speechify, ElevenLabs or OpenAI with your keyNo, reads the replyNot listedDepends on the voiceFree; paid voices on your own keyYes (MIT)macOS, Windows, Linux; Claude Code plugin and a VS Code extension
claude-voice-notifyLocal, the system voiceYes, a line Claude writes at the end of its replyYes, says the repository nameEnglish, Brazilian PortugueseFreeYes (MIT)macOS (Linux and Windows new, not tried on real machines)
PintailCloud (OpenRouter or Pintail credits); Kokoro on the Mac for EnglishYes, four levelsYes: a notch pill with live captions, and a desktop duckEnglish, German; five more in beta$15 once + ½–2 ¢ a replyNomacOS 15 or later, Apple silicon

Checked against each project’s own page on September 30, 2026. Heard, Agents Aloud, claude-speak, Kobi and the Pintail row were checked on October 29, 2026. Heard’s /say skill (claude-code-tts), spoken-replies (claude-voice), claude-read-aloud and claude-voice-notify were checked on November 4, 2026. Tell me if something changed: I’ll fix it.

A DIY Stop hook

Claude Code runs a hook when a reply finishes. Pipe the text into the system voice and you have speech in about 20 lines. This walkthrough shows how. It reads markdown literally, stars and backticks included, tabs talk over each other, and it speaks up when a background job has only just started.

Pick it if you like tinkering and want something free today.

macOS Spoken Content

Built into every Mac: select text, press a shortcut, the system voice reads it. No setup beyond a checkbox, but it reads whatever you select, all of it, and you have to select it.

Pick it if you only want the odd reply read and don’t mind the system voice.

Aloud

Open source (MIT) and fully local: Kokoro speaks on your Mac. It works with Claude Code and Codex, gives a short gist of each reply with the full reply on request, keeps sessions apart, and speaks up when an agent asks a question or needs approval. You install it with Homebrew, Python and pipx.

Pick it if you want it free and fully local. Pintail can speak English on the Mac with Kokoro too; its other languages use the cloud.

VoiceMode

An open-source (MIT) MCP server for two-way voice conversations with Claude Code: you speak, it answers out loud. Local Whisper and Kokoro, or cloud services. It runs on macOS, Linux and Windows. I tried it before building Pintail and removed it, because I already had dictation I liked; I wanted the replies read, not a conversation.

Pick it if you want to talk to your agent hands-free, or you’re on Linux or Windows.

SpeakySpeak

Open source (MIT) and fully local: a menu bar app that reads each Claude Code reply with Kokoro, or the system voice on Intel. It’s built for parallel sessions too: replies queue and play one at a time, you can mute a session, and it says out loud when a session is waiting for your permission. It reads the whole reply, cleaned up for the ear, not a summary. The neural voice is English only. Claude Code builds it from source on your Mac; macOS 15.6 or later.

Pick it if you run several Claude Code sessions, work in English and want it free and local.

claude-voiceover

An open-source (MIT) Claude Code plugin that reads the final reply with your system voice, fully local. It skips code blocks, tables and links, lets you switch speech off for one session with /voiceover off, and has an optional menu bar app for the voice and speed. Runs on macOS and Linux.

Pick it if you want a free plugin that installs in a minute and the system voice is fine for you.

flowcode

A native macOS menu bar app, open source (MIT) and in public beta. It reads Claude Code replies aloud with a local voice and does hold-to-talk dictation too, both on your Mac, in English and Czech.

Pick it if you want reading and dictation in one local app, or you speak Czech.

Heard

Heard is a Mac app for hearing coding agents. The engine is Apache 2.0 and speaks with local Kokoro; you can build it and bring your own keys. The downloadable app is their managed build. On the pricing page, the free plan is $0, and Power is $24 a month billed yearly ($288 a year), with a 14-day trial and no card. It needs macOS 14 or later. Hooks cover Claude Code and the Codex CLI. That page also lists Cursor, GitHub Copilot CLI, Gemini CLI and others for sessions and approvals, in any terminal, including Ghostty, iTerm2 and tmux. Power adds managed voices and a model, up to 2 million characters a day, spoken recaps, “Hey Parrot,” and an iPhone companion. Languages are not listed there.

Pick it if you want a free local engine, or you’ll pay monthly for managed voices and recaps.

Agents Aloud

Agents Aloud is a personal macOS app, MIT, that you build from source. It needs macOS 26 or later to run and Xcode 26 or later to build. It reads Claude Code, from a live process list, and Codex transcripts, and it only reads. One shared queue speaks the replies. When it moves to another session it says “From” and the session name; with one session it stays quiet about the name. Voices are the Mac’s Spoken Content voice, or ElevenLabs with your own key. An optional rewriter can run through the Claude or Codex CLI, and it’s off by default. With the default voice and the rewriter off, the README says nothing leaves the machine.

Pick it if you want a free, read-only queue on macOS 26 and you’ll build it yourself.

claude-speak

claude-speak is an MIT Claude Code plugin. Kokoro speaks on your machine, and a one-line summary is written with claude -p on Haiku, using your Claude login. Turn summaries off and, after the voice download, it needs no network. It speaks each finished reply. Several sessions take turns, and each reply starts with the project name. You need uv and an audio player (afplay on a Mac). It runs on macOS and Linux. The first download is about 335 MB. The pitch is English summaries; Kokoro includes other voices. Install with /plugin marketplace add kyleoliveiro/claude-speak.

Pick it if you want a free plugin and a one-line English summary, and you’re fine using your Claude login for that line.

Kobi

Kobi is an MIT Claude Code plugin. Kokoro speaks on the Mac by default, and those lines stay there. ElevenLabs is optional; their README’s example is about $8 a month at about 100 lines a day. It can also fall back to the Mac’s system voice. It speaks when Claude needs an OK, is stuck, fails, or finishes, with a one-line summary, and it names the project. Two sessions in one project get the session name too. It works in the terminal and in the Code tab of the Claude Mac app, and only while /voice is on. Kokoro needs Apple silicon, macOS 15 or later, the Command Line Tools, and about a 460 MB voice. An optional menu bar app needs macOS 26 and Apple silicon; the plugin does not. Install with /plugin marketplace add sdusteric/kobi.

Pick it if you want a free local voice that speaks when Claude is stuck or finished, and names the project.

Heard’s /say skill (claude-code-tts)

claude-code-tts is a free MIT repo from the Heard team. It has two parts: a Stop hook that pipes last_assistant_message into a script, which strips markdown, caps the reply at 600 characters and reads it with the system voice, and a /say skill that speaks a short summary when you ask for one. Parallel sessions queue instead of talking over each other. It runs on macOS, Linux and Windows, with Claude Code and Codex.

Pick it if you want the DIY hook already written, plus a summary on demand, and the system voice is fine.

spoken-replies (claude-voice)

spoken-replies is an MIT Claude Code plugin for macOS. While it’s on, a hook tells Claude that the reply will be read aloud, so Claude writes the whole reply for the ear, and the system voice reads it. Opening your microphone stops the speech, it pauses music and videos while it talks, and a ding plays when Claude needs permission or asks a question.

Pick it if you dictate a lot, run one session at a time, and don’t mind Claude changing how it writes the reply.

claude-read-aloud

claude-read-aloud is an MIT Claude Code plugin plus a VS Code extension with a speaker button in Claude’s chat. Voices go from free system voices and local Kokoro to Speechify, ElevenLabs or OpenAI with your own key. It reads the reply rather than a summary, and auto-read is off by default.

Pick it if you use Claude Code inside VS Code and want a play button and your choice of premium voice.

claude-voice-notify

claude-voice-notify is an MIT plugin, one Python script with no dependencies. When a session finishes it says the repository name and a short line that Claude writes at the end of its reply, and it tells you when a session waits for approval. It uses the system voice, in English or Brazilian Portuguese. The author uses it daily on macOS; Linux and Windows are new and not yet tried on real machines.

Pick it if you want a free spoken “done” with the repo name and the system voice is enough.

Pintail

A Mac app that turns each reply into a short spoken summary at the length you pick (one sentence up to the whole reply). The voice runs in the cloud on your own OpenRouter account or on Pintail credits, and English can use Kokoro on the Mac. It’s built for many tabs: one voice at a time, and a pill with the project name and live captions, plus a desktop duck, so you know which session finished. English and German. Spanish, French, Italian, Dutch and Portuguese are in beta. $15 once. A cloud reply costs usually ½–2 ¢ a reply.

Pick it if you run several Claude Code sessions at once, want a short summary with the project named in the notch, and want to pay once. How it works and pricing have the details.

What about Claude Code’s /voice?

That’s dictation: you speak your prompt, Claude Code types it. It doesn’t read replies back. Claude Code has no built-in read-aloud, which is why all of the above exist. More in Claude Code voice mode.

Try Pintail

20 free read-outs, then $15 once. No subscription.

Download free trial