Codex Ditches Voice in the Terminal
We break down Codex’s removal of WebRTC voice and dictation from the TUI, why the team is pushing those workflows to the desktop app, and what terminal users can do instead. Plus: the latest alpha hotfixes, including a rollback for tilde-based URL parsing and cleaner plaintext handling for web search context.
Chapter 1
Codex Strips Built-In TUI Voice Support
Ethan Park
Welcome to the show everybody! I'm Ethan Park, here with Maya. And first off, we have to give a quick shout-out and thank you to Jellypod for supporting the show. Now, Maya, let's look at something that caught my eye in the latest Codex upstream cleanup. PR number 27801. They completely ripped out the WebRTC-based real-time voice and dictation system from the Codex TUI. It's just gone. [surprised]
Maya
Wait, [questioning tone] they completely deleted it? So the actual `/realtime` session command and all those audio options in `/settings` are just... wiped from the terminal interface?
Ethan Park
Completely wiped. [matter-of-fact] If you try to run `/realtime` now, you'll get an error. The main driver here was streamlining the binary footprint. WebRTC brings a massive amount of overhead, and shipping those heavy media libraries inside a terminal-focused CLI tool was starting to look like a massive case of feature creep.
Maya
I mean, [pauses] that actually makes a lot of sense. WebRTC is notorious for bloating binaries. But if you're a developer who actually used native terminal-first dictation with your own API keys, this is a pretty abrupt breaking change. What are they supposed to do now?
Ethan Park
The official recommendation is to shift those workflows over to the visual desktop application—the `codex app`—which is better suited to handle heavy media pipelines. Or, if you absolutely must stay in the terminal, you have to route your voice input through native operating system dictation utilities instead.
Maya
Right, [thoughtfully] so instead of Codex handling the microphone stream, you let macOS or Windows do the heavy lifting of speech-to-text at the system level, and just feed the plaintext into the terminal. It's less integrated, but honestly, it keeps the terminal client lean.
Chapter 2
Pre-Release Hotfixes and Terminal URL Restorations
Ethan Park
Exactly. And that brings us to the latest 0.140.0-alpha.15 hotfixes, which really show this strategic shift toward tightening the CLI's core competency around raw text and tool execution. For starters, they had to roll back a buggy URL parser that broke clickable terminal paths containing tildes. [deliberate]
Maya
Oh, the tilde regression! [chuckles] That must have been incredibly annoying. If you can't click a relative path like `~/projects/src` because the parser doesn't recognize the tilde, your terminal-to-editor navigation completely breaks.
Ethan Park
Precisely. It was a major papercut for developer velocity, so rolling that back in alpha 15 is a welcome fix. Along with that, they've also improved how code-mode web search handles nested tool calls. It now returns highly optimized plaintext context.
Maya
Ah, [curious] so instead of dumping raw HTML or heavily formatted markdown that bloats the LLM's context window, it strips all the noise out?
Ethan Park
Yes, exactly. It prevents context window bloat by keeping the payload strictly to clean, relevant plaintext. It really shows where their focus is. They are offloading the heavy media components like voice to the desktop app, and hyper-focusing the terminal CLI on what it does best: fast, low-overhead text manipulation and precise tool execution.
Maya
It's a classic Unix philosophy move. [calm] Do one thing and do it well. Let the desktop app be the heavy, multi-modal playground, and keep our terminal fast and dependable.
Ethan Park
Couldn't agree more. That is all the time we have for this quick take. Thanks for listening, and we'll catch you in the next one. [warmly]
Maya
See ya! [warmly]