Speech to text doesn’t work

[Samigos]

I've downloaded all models, and every time I record something, the same text comes out... "you". I think that the app doesn't have access to the right mic, or something. (I only have 1 mic though)

I'm on macOS 15.1 and CodexMonitor 0.7.3

Screen.Recording.2026-01-19.at.17.37.16.mov

[Dimillian]

CC @moehaje
Working here FYI

[moehaje]

I’ll take a look. This sounds like it may be something local on the machine (mic input/permission), so could you double-check CodexMonitor has Microphone permission enabled in macOS Settings?

[moehaje]

Can you maybe run the app from the dev server and reproduce it? Would be great if you could paste the terminal output lines that start with dictation: from the moment you hit record to when you stop

[radu2lupu]

The problem is that the app does not request microphone access. I don't see it on the list to even enable it.

[Dimillian]

@codex can you check when we check for the microphone permission, if it broke before from one of the commits and fix it? It should be asked when trying to use the feature from the composer (when the model is downloaded, ready) and the user press the microphone from the composer just before we start the recording.

[chatgpt-codex-connector]

Summary

• Added a macOS microphone permission request flow (including denial messaging) before dictation starts so the prompt happens right before recording begins. src-tauri/src/dictation.rsL21-L758
• Added macOS AVFoundation/objc dependencies needed to trigger the permission prompt. src-tauri/Cargo.tomlL46-L51

Testing

• ⚠️ cargo check (failed: missing glib-2.0 system library in this environment)
• ✅ npm run lint
• ✅ npm run typecheck

View task →

[radu2lupu]

Fixed: #129

[radu2lupu]

Or hey, too late eh. Your LLM beat my LLM in allowing voice dictation to LLMs

[Dimillian]

Merged yours haah