Voice Input & Dictation

Talk to your agents — on-device Whisper transcription by default, with an optional cloud mode for speed and accuracy.

Voice input lets you talk to your agents instead of typing. A mic button in the composer records what you say and inserts the transcription straight into the message. By default everything runs on-device, so nothing leaves your machine.

How it works

Look for the mic button in the composer. Tap it to start recording, speak your prompt, and stop when you’re done — Dices transcribes the audio and drops the text into the composer, where you can edit it like anything you’d typed.

1. Place your cursor in the composer (start a new message, or add to one you’re drafting).
2. Click the mic button.
3. Speak your prompt naturally.
4. Stop the recording.
5. The transcription appears in the composer — review, edit, and send.

Tip: Dictation is great for the long, rambly setup prompts you’d never bother to type — describe the bug, the feature, and the constraints out loud, then clean up the transcript before sending.

Local mode (default — no key)

Out of the box, Dices uses local mode. Whisper runs on-device through Transformers.js, so transcription happens entirely on your own machine.

• Private by design — your audio never leaves your computer.
• No API key required — it just works.
• First-use download — a small Whisper model is downloaded the first time you use voice input. After that, it’s available offline.

Note: Because local mode downloads a model on first use, the very first transcription may take a moment while the model is fetched. Subsequent transcriptions are local and don’t need that step.

Cloud mode (optional)

If you want faster or more accurate transcription, you can switch to cloud mode by adding an API key under Settings → Voice. Dices supports:

• OpenAI — add an OpenAI API key.
• Groq — add a Groq API key.

With a key configured, your recorded audio is sent to that provider for transcription, and the result comes back into the composer the same way.

Setting up cloud mode

1. Open Settings → Voice.
2. Add your OpenAI or Groq API key.
3. Use the mic button as usual — transcription now runs through that provider.

Warning: In cloud mode your audio is sent to the provider you configured (OpenAI or Groq). If you need transcription to stay entirely on your machine, use local mode instead — it’s the default and sends nothing out.

Local vs cloud at a glance

Next steps

• Privacy & Security — the full picture of what runs locally and what leaves your machine.
• Commands & the Palette — slash commands and @-mentions to pair with your dictated prompts.