Codex CLI voice is now switched on by default, so the next time you open a session, your terminal can listen to you. The change landed in version 0.156.0 on September 22, 2026, and you toggle the mic with a single key. If you work on shared machines or CI, you’ll want to know how to turn it off.
Here’s the short version of what matters before the details.
- Version: Codex CLI 0.156.0 made voice a default, per a September 22 write-up.
- Toggle: press F8 to start or stop listening.
- Safety net: Codex waits for you to confirm the transcript before acting.
- Opt out: set
voice.enabled = falsein your config file. - Latest build: version 0.161.0 (October 7) adds microphone and speaker pickers.
The list covers the essentials. The sections below explain how each piece works in practice.
What “voice on by default” actually means
Before 0.156.0, you had to opt in with a flag or a config change. Now the voice runtime starts with every session. Nothing records you until you press F8, though, so it isn’t an always-listening mic.
OpenAI also called out voice control in its DevDay 2026 recap, listing a refreshed Codex CLI with voice control and an /agents view. If you read our DevDay preview, this is one of the pieces that shipped.
The practical effect: you can say “run the tests and show me what failed” while your hands stay on something else. Then you read the transcript and confirm it.
How to use and tune Codex CLI voice
Using it takes about ten seconds to learn. These are the controls reported for the release.
| Control | What it does |
|---|---|
| F8 | Toggles listening on and off |
| /voice settings | Opens the language and microphone picker |
| voice.enabled = false | Turns voice off in your config, meant for headless setups |
| /usage | Shows account-level spend and token totals |
Per the write-up, Linux and Windows audio runtimes now ship bundled. That means no separate install to get sound working.
Where voice helps, and where it doesn’t
Voice works best for short, well-defined requests. Think “open the failing test,” “summarize this diff,” or “start a new session on the login bug.” These are quick to say and easy to check on screen.
It works less well for precise edits. File paths, variable names, and regular expressions are awkward to speak, and speech-to-text can mangle them. For those, typing is still faster and safer.
A good habit is to mix both. Speak the intent, then fix the details by keyboard before you confirm. That keeps the speed of voice without the risk of a misheard command.
Why you might want to turn it off
On a shared terminal or a CI runner, there’s no microphone to capture. A voice runtime that starts anyway can throw errors or waste startup time. Setting voice.enabled = false fixes that.
It’s also worth turning off in open offices. A spoken command can leak what you’re working on, and a transcript you skim too quickly is easy to confirm by mistake.
The other changes in 0.156.0
Voice gets the headline, but the release has a few other upgrades that make daily use smoother.
- The fullscreen
/tuimode is now stable, with transcript search, mouse selection, and right-click copy. - Worktrees are on by default, so parallel agent sessions stay in separate branches of your files.
- Mermaid diagrams and equations render right in the terminal.
- Six new themes cover light, dark, and high-contrast looks.
Worktrees matter most if you run several agents at once. Each one gets its own working copy, so they don’t overwrite each other’s edits.
What’s changed since: version 0.161.0
The Codex CLI releases page lists version 0.161.0 as the latest, dated October 7. It lets you choose your microphone, speaker, and input channels for voice conversations, and it saves those choices locally. It also adds /mcp login <name> for signing in to MCP servers from an active session.
If you pay through ChatGPT, our guide to Sign in with ChatGPT plan usage explains how your plan limits carry into tools like this one.
Frequently Asked Questions
Is Codex CLI always listening to me?
No. The voice runtime starts with the session, but listening begins only when you press F8. Codex also waits for you to confirm the transcript.
How do I disable voice in Codex CLI?
Set voice.enabled = false in your config file. This is the recommended route for shared terminals and CI.
Which version added voice by default?
Version 0.156.0, released September 22, 2026, according to a third-party write-up. Check the official release notes for your install.
Do I need to install an audio runtime?
Not on Linux or Windows. The reported release bundles those runtimes, so voice should work without extra setup.
Our take
Voice in a coding terminal sounds gimmicky until you use it for short, clear commands like “rerun the last test.” Leave it on for your own machine, since the confirm step keeps it safe. Turn it off anywhere shared or automated, and update to 0.161.0 to get the device pickers.
Leave a Reply