Voice chat

A hands-free spoken conversation with your coding agent, in the iOS app. You talk, it talks back, and it keeps listening. Not tap-to-dictate, an actual back-and-forth you can run while your hands are on the wheel or the keyboard. The whole voice layer is native and on-device, so nothing about your speech leaves the phone.

What it is

Voice chat lets you hold a spoken conversation with the agent driving your codeout session. You speak a request, it speaks a reply, and the loop continues until you close it. This is iOS only, and it is built on Apple's on-device speech, not a dictation field you have to keep tapping. You drive the agent with your voice the way you would drive it by typing.

Fully native and on-device

Speech-to-text on the way in and text-to-speech on the way out both run on your iPhone. No cloud service, no API keys, no extra cost, nothing transmitted for the voice layer. It is the same private, self-hosted ethos as the rest of codeout. The daemon is untouched by any of this: it just sees a normal text turn arrive and sends a normal text reply back, exactly as if you had typed.

Note

Because the voice layer is entirely on the device, your microphone audio is never uploaded anywhere. The agent on your daemon only ever receives text.

What makes it a real conversation

This is more than dictation. It is a continuous loop that handles the timing for you, so it feels like talking rather than operating a recorder.

Listen, speak, listen

A continuous loop. After the agent finishes speaking, it goes straight back to listening, so you can keep the conversation going without touching the screen.

Automatic turn-taking

It sends when you pause. There is no Send button to hit; a natural break in your speech ends your turn and hands it to the agent.

Barge-in

Start talking while it is speaking and it stops to listen. You can cut in mid-sentence to redirect, the way you would with a person.

Streaming speech

It talks back sentence by sentence as the reply arrives, rather than waiting for the whole turn to finish before it says a word.

Coding-smart speech

Reading a coding agent aloud verbatim would be unbearable. Voice chat speaks the assistant's prose and summaries and skips the code blocks, diffs, and tool output. So you hear "tests pass, fixed the null check in auth.swift" instead of two hundred lines of source read out one character at a time. The full reply is still there on screen to read; the voice just narrates the parts that make sense out loud.

Using it

Tap the waveform button in the composer to open the full-screen voice conversation. It needs microphone and speech-recognition permission, which iOS asks for the first time. From there, just talk: pause to send, listen to the reply, and talk again. Close the voice view to go back to the normal chat at any point.

Honest note

Under the hood this is speech-to-text plus text-to-speech, like every voice mode for a text agent. There is no separate voice model. The difference is that the on-device voices are good and the whole thing is completely private, since none of the audio ever leaves your phone.