# Voice input

> Dictate messages to the KazoCode agent. Speech is turned into text in real time on your computer with local Whisper, or with OpenAI, Mistral or Windows speech recognition.

Talk to the agent instead of typing. Click the microphone next to the send button, speak, and your words appear in the message box as you talk.

## Local Whisper

When you haven't added an OpenAI or Mistral key, speech is recognized on your computer with Whisper, so your voice never leaves it and it works offline.

The Whisper model isn't part of the KazoCode download. The first time you use the microphone, KazoCode asks whether to download it and shows the progress. Nothing is downloaded until you agree.

While you speak, the phrase so far is transcribed about every 0.6 seconds, so words appear almost as you say them; a pause settles the phrase.

## Choosing the engine

With an OpenAI or Mistral API key in [Models and providers](https://kazocode.app/docs/agent/models.md), KazoCode uses that provider's speech recognition. Without one, it uses the engine you choose in **Settings › General › Voice Input**:

| Engine | Where speech is recognized |
| --- | --- |
| **Local Whisper** | On your computer (the default) |
| **Windows Speech** | By Windows' built-in speech recognition |
