Skip to content

BrokenCode guide

Local dictation in BrokenCode

Download a speech model once, dictate on your device, and review the transcript in your BrokenCode draft.

UPDATED 01 Oct 20263 min readTECHNICAL GUIDE
01

Press the microphone. Keep the words.

  1. Update to BrokenCode 1.18.45 or later for improved English dictation and configurable keyboard shortcuts.
  2. Press the microphone in the desktop composer. The first use downloads the local speech model and shows progress. Allow microphone access when your operating system asks.
  3. Speak while the waveform responds to your voice. Press Finish dictation to transcribe on your device.
  4. Review the text in your existing draft, then send when ready. Dictation preserves attachments and does not send the message automatically.
02

Dictate from the keyboard.

With BrokenCode focused, press Command + Shift + M on macOS or Control + Shift + M on Windows and Linux to start dictation in the current message. Press the same shortcut again to finish and transcribe. Escape cancels the active recording, model download, or transcription.

Open Settings → Shortcuts and find Start or finish dictation or Cancel dictation to assign your preferred keys. The shortcuts use the same microphone, progress, and local transcription controls as the composer button. Review the transcript before sending.

03

Download once, load automatically.

macOS, Windows, and Linux use a larger English speech model for improved dictation. The first microphone click downloads about 2.42 GB and verifies the weights. Interrupted downloads can resume. The desktop app keeps weights in its own local cache and warms them on later launches. Audio is transcribed on your computer. Browser dictation uses a separate local English model and browser cache.

Record up to two minutes at a time on desktop. Cancel stops the microphone and discards that recording. Loading and transcription can also be cancelled. Microphone permission failures and model download errors show a retryable message.

04

Native speech on your phone.

Android first checks for advanced local speech, currently supported on Pixel 10 and Pixel 11. Other supported Android 12 or later phones use the system on-device recognizer. English packs download when needed, and supported services can show progress. The system decides language availability and may schedule a download. Quality punctuation and coding vocabulary are requested on supported services.

Supported iOS 26 phones use the newer local speech transcriber with automatic English asset downloads. Earlier supported phones use local English dictation with punctuation and coding vocabulary. Text joins the current draft for review. Remote workspace access continues to work on supported phones without dictation.

05

Small cues, clear controls.

The microphone, waveform, progress, and cancellation controls show the current state. Desktop animations respect the system’s reduced-motion preference. Speak naturally, finish the recording, and edit the transcript before sending.

STRAIGHT ANSWERS

Frequently asked questions

01Is my voice uploaded for transcription?

No. Desktop recordings are transcribed by the downloaded local model. Native phone dictation requires an on-device speech service.

02Does the model stay on disk?

Yes. Desktop weights remain in the app’s local speech cache between launches. Removing that cache requires another download. Browser dictation keeps its separate weights in browser storage.

Download

Say the next instruction.

Update BrokenCode and use the microphone in your composer.

Download BrokenCode