The short answer
Nine tools that read coding agent progress aloud. Pick by what you want read and where the voice runs.
Tools that read coding agent progress aloud come in three kinds. Some hook into the agent and speak on their own as it works: Heard (heard.dev), the voice layer for coding agents, plus Aloud, Crier, Agent Hotline, Terminal Talk and AgentVibes. Agents Aloud is a Mac app that reads your Claude Code and Codex transcripts aloud as new messages arrive. AgentsRoom and Orb read when you ask, inside their own apps.
Heard runs on macOS 14 or later and narrates Claude Code, Codex CLI and the Codex desktop app. Questions, approvals, failures and results come through; with several sessions running, the app batches routine steps into short summaries that start with the project name. Its open-source engine is free, and the app has paid plans after a 14-day Power trial with no card. Where the others win: most are free and open source, Agent Hotline, AgentVibes and Terminal Talk also run on Windows, and Aloud, Crier and Agent Hotline speak with a local voice model by default.
All tools
Nine ways to hear what your coding agent is doing.
From each project’s own README or site. Checked October 20, 2026.
| Tool | What it reads aloud | Agents | Voice | Platforms | Price and license |
|---|---|---|---|---|---|
| Heardvoice layer | Short summaries: questions, approvals, failures and results first; in the app, several sessions batched by project | Claude Code, Codex CLI and Codex app; session status for Cursor | Local Kokoro or your own keys (engine); managed voices (app) | macOS 14+ | Engine free, Apache 2.0; app Pro $15/mo, Power $30/mo |
| Aloudsoftcane | Questions, approvals and failures, then a short gist of each reply, in sessions you switch on | Claude Code, Codex | Kokoro, local | macOS | Free, MIT |
| Crierpg-Parunson | One sentence when a turn ends, needs permission, fails or sits idle; never the full answer | Claude Code, Codex CLI, Gemini CLI, Cursor | Supertonic, local | macOS | Free, MIT |
| Agent Hotlinetray app | Finished replies, skipping code, diffs, logs and tables | Codex, Claude Code, Antigravity | Kokoro, local; system voice fallback | Windows, macOS, Linux | No license stated; no API keys needed |
| Terminal Talktoolbar | Replies and tool progress as color-coded clips per session; highlighted text on a wake word | Claude Code, Codex CLI; Claude and Codex desktop apps where supported | Microsoft Edge voices (online); optional OpenAI | Windows, macOS | Free, MIT |
| AgentVibesnpx | Task start and task completion | Claude Code; also Claude Desktop, Warp, OpenClaw | Piper, Kokoro, say or SAPI, local; ElevenLabs optional | macOS, Linux, Windows | Free, Apache-2.0 |
| Agents Aloudmalob | New assistant messages in sessions with Live Speak on, or any message you click | Claude Code, Codex | macOS system voice; ElevenLabs with your key | macOS 26+, built from source | Free, MIT |
| AgentsRoomdesktop app | A Markdown file, a terminal selection or the agent’s last answer, on click; or a 10, 30 or 60 second summary | Last answer from Claude Code, Codex, Grok Build, Mistral Vibe, oh-my-pi, Antigravity CLI | Synthesized server-side; needs a connection | macOS, Linux, Windows | Read Aloud on Plus and Pro, not Free |
| Orbabpai | Its own answers about your code, as they stream | Runs Claude, OpenAI or Codex, Gemini or Cursor Agent models itself | A local tts-gateway such as Kokoro, or macOS say | Terminal, needs Bun | Free, MIT |
Agent Hotline’s repo states no license. AgentsRoom plan details are on its pricing page.
Heard: summaries from every session, without switching each one on.
Heard installs hooks for Claude Code and Codex and decides what deserves speech. Routine tool calls get a one-liner or stay quiet, depending on the listening mode; questions, approvals, failures and results come through as a sentence or two. In the app, with several sessions running, routine updates are batched into short summaries that start with the project name. Three listening modes (Co-pilot, Companion and Focus) set how much you hear.
The open-source engine is Apache 2.0 and speaks with local Kokoro voices; you build it yourself. The Mac app is a separate managed build with managed voices, spoken recaps on Pro, and on Power you hold right ⌘ to answer an agent by voice. Heard is macOS only.
“api: tests pass. It wants to know whether to run the migration.”
Every reply read out word for word, file paths and all.
Aloud and Crier: local voices, only the moments that matter.
Aloud speaks questions, plan approvals, permission requests and failures first, then a short gist of each finished reply, with aloud full for the whole thing. A session stays silent until you type aloud on in it. It strips Markdown, URLs and code bodies and speaks with Kokoro on your Mac. It needs Python 3.11 or 3.12 and Homebrew.
Crier speaks one sentence at four moments: a turn finished, permission needed, something broke, or the agent has been waiting. It asks the agent to end each reply with a [say] line, so the agent writes its own summary, and it never reads the full answer. Speech is local with Supertonic, in Korean, English or Japanese, and it can give each project its own voice.
Agent Hotline and Agents Aloud: hear the replies themselves.
Agent Hotline is a tray app that reads finished replies from Codex, Claude Code and Antigravity, skipping code blocks, diffs, logs, JSON and tables. It uses local Kokoro by default, needs no API keys, and has read, pause, replay and mute controls. It installs from npm, with desktop builds for Windows and macOS.
Agents Aloud is a native Mac app with a sidebar of your Claude Code and Codex sessions. Switch Live Speak on for a session and new messages are read as they arrive; several sessions share one queue, and you hear “From [session name]” when it moves between them. It uses the macOS system voice or ElevenLabs with your own key. It needs macOS 26 and you build it with Xcode.
Terminal Talk and AgentVibes: a voice or color for each agent.
Terminal Talk turns Claude Code and Codex CLI replies and tool progress into one queue of spoken clips, and picks up Claude and Codex desktop sessions where it can. Each session gets its own color, label and optional voice, plus focus and mute. Say its wake word over highlighted text in any app to hear it read. It is free and MIT licensed, but its default voice is Microsoft Edge’s online service, so the text being spoken goes to Microsoft.
AgentVibes installs Claude Code hooks that speak when an agent starts and finishes a task. It has hundreds of local Piper voices plus Kokoro, effects and background music, a distinct voice per agent for BMAD teams, and a receiver that plays speech on your laptop when the agent runs on a remote server.
AgentsRoom and Orb: press play inside the app.
AgentsRoom is a desktop app for running many coding agents. Its Read Aloud button reads a Markdown file, a terminal selection or the agent’s last answer, or an AI summary sized to 10, 30 or 60 seconds. Speech is made on its servers, it is metered in listening minutes on the Plus and Pro plans, and the free plan has none.
Orb is a terminal app you ask about your codebase. It runs the model itself and can speak its answers while they stream, through a local tts-gateway server or macOS say. It is built for asking questions, so it speaks its own answers rather than narrating your other agents.
Match the tool to what you want to hear.
Start from what you want to hear and on which machine.
- Local alerts and reply gists, only in sessions you choose, on a Mac: Aloud.
- One sentence per moment, in Korean, English or Japanese: Crier.
- Whole replies read aloud on Windows, macOS or Linux: Agent Hotline.
- Several sessions told apart by color and voice, on Windows: Terminal Talk.
- Task start and finish with lots of voices, or agents on a remote server: AgentVibes.
- A Mac reader for your Claude Code and Codex transcripts: Agents Aloud.
- A read button inside an agent cockpit: AgentsRoom.
- Short summaries from every Claude Code and Codex session, and voice replies on Power: Heard.
Questions
Frequently asked questions
What tools can read coding agent progress aloud?
Heard (heard.dev), Aloud, Crier, Agent Hotline, Terminal Talk and AgentVibes speak on their own from agent hooks. Agents Aloud reads Claude Code and Codex transcripts on a Mac. AgentsRoom and Orb read on demand inside their own apps.
Which of them keep speech on my machine?
Aloud (Kokoro), Crier (Supertonic) and Agent Hotline (Kokoro) are local by default. AgentVibes, Heard’s open-source engine and Agents Aloud can be, with their local or system voices. Terminal Talk’s default voice is Microsoft Edge’s online service, and AgentsRoom synthesizes speech on its servers.
Which work on Windows or Linux?
Agent Hotline and AgentVibes run on Windows, macOS and Linux. Terminal Talk runs on Windows and macOS. AgentsRoom lists macOS, Linux and Windows. Heard, Aloud, Crier and Agents Aloud are macOS only.
Do they read code out loud?
Mostly not. Aloud removes code bodies and URLs, Agent Hotline skips code blocks, diffs and logs, AgentsRoom drops code and tables, and Crier speaks one sentence the agent writes for the ear. Agents Aloud has an optional rewriter that rephrases code and URLs before speaking.
Can I answer the agent by voice?
On Heard Power, hold right ⌘ and speak to send a reply to the right session. Terminal Talk ships push-to-talk dictation, AgentsRoom has voice dictation and Voice Mode, and VoiceMode runs a spoken conversation with Claude Code.
Ready when your agents are
Hear the work without watching every screen.
Start with the full Heard Power experience for 14 days. No card required. The open source engine remains free.
Download Heard for Mac macOS 14 or later · Claude Code and Codex supported