Read Aloud
- Desktop, mobile and web
- Plus, Pro
What it does
Reads written text out loud in a natural synthetic voice: a markdown file, a terminal selection, the agent's last answer, or an AI summary. Every entry point offers two modes: Read it all, word for word, or a condensed read where an AI summary is written first at the spoken length you pick (about 10, 30 or 60 seconds) and that summary is what you hear. The text is cleaned before speaking: code blocks, markdown tables, bare file paths, rules and markup are dropped so you hear prose, not punctuation. Built for a dyslexic developer who was pasting the preview into an external reader; also handy at the end of the day for a four-paragraph answer you would rather hear while typing.
Where to find it
| Surface | Control | What is read |
|---|---|---|
| Markdown file preview | speaker button next to Source | the whole file |
| Terminal selection panel | Listen | the highlighted text |
| Terminal right-click menu | Read the last answer / Listen to the last answer | the agent's last prose answer, from its transcript |
| AI summary or explanation bubble | speaker button | that result |
| Agent pane floating controls (optional) | Listen to the last answer (n s) | the last answer, at your default level |
While reading, a control chip in the status bar shows progress, remaining time, pause and stop; it is the control that stays available after you close the file preview.
How to use it
- Open a README, CLAUDE.md or changelog in the preview and click the speaker, or select text in an agent terminal and click Listen.
- Pick a level: Read it all, Flash summary, Medium summary, Long summary. One click on the speaker uses your default level; the others stay one click away.
- Listen. Pause, resume or stop from the button or the status-bar chip. Closing the preview does not stop the read.
- Starting another read stops the current one: one reader for the whole app.
Settings
readAloudDefaultMode(global, Settings > Composer > Read aloud > Level used on a single click):full,flash(default),mediumorlong.readAloudPaneButton(global, same section): show the listen button in the floating controls of agent panes.stopMusicWhenReading(global, same section): pause the music while reading.voiceModeVoice(Voice Mode panel > Reply voice): the voice used by both Voice Mode and Read Aloud.
Agent tools (MCP)
None.
Providers
Reading a file, a selection or an AI summary works with every provider (the text comes from your screen). Reading the agent's last answer needs an addressable transcript on disk, which today means Claude Code and Codex; other providers report there is nothing to read.
Mobile
Present on the mobile companion with the same levels, the same character caps and the same plan gate: Listen to the last answer on an agent, a control chip while reading, and the default level in the phone's Settings > Read aloud. The phone uses the server default voice (no voice picker yet) and plays even in silent mode.
Limits
- Plan gate: no included listening on Free; the plan includes a monthly listening allowance on Plus and Pro (upsell shown otherwise, and in the web demo). Beyond the allowance, reads spend voice credits, shared with dictation and Voice Mode; the balance is checked before anything is spoken. Your own OpenAI key removes the spend.
- Hard cap per read: about 12,000 speakable characters; past that the read is truncated and marked "shortened".
- The condensed read spends two resources: the monthly AI summaries quota for the text, then voice credits for the audio.
- A document is read in the language it is written in; only the summary is written in your interface language.
- Starting a read closes Voice Mode (its open mic would hear the voice).
Common questions
- Does it read code? No, deliberately: code blocks, tables and paths are removed first.
- Can I choose the voice? Yes, it is the Voice Mode reply voice, set once in the voice panel.
- What happens if I close the file while it reads? The read continues; stop it from the status-bar chip.
- Why "shortened" on a long changelog? The per-read cap protects your credits; use a condensed level for long documents.
- Why can't it read my agent's last answer? That provider has no readable transcript; select the text in the terminal and click Listen instead.
Related
- Voice Mode: the conversational counterpart, same voice and credits.
- Voice Dictation: speech in, where Read Aloud is speech out.
- File Preview: the markdown preview that carries the speaker button.
- Bring Your Own OpenAI Key: reads on your own key.