Dictate Your Prompts with Voice Input
You can speak your prompts to Mythex instead of typing them. Click the microphone or press Alt+V, check the text, and send. How it works and its limits.
Mythex Team · · 3 min read
You can talk to Mythex instead of typing. The composer has a microphone button: click it (or press Alt+V, Option+V on a Mac), say what you want, and your words appear as text in the message box. You read it over, fix anything that came out wrong, and send it like any other prompt.
What it does
Voice input turns speech into the text of your prompt. It uses your browser's built-in speech recognition, so there is nothing to install and no separate app to open.
It works in the composers that use Mythex's shared prompt bar. That means the chat inside a project workspace, and the home composer where you start a new project, since it uses the same control.
The transcribed words land in the message box. Nothing is sent until you send it, so you always get a chance to read what was heard before the agent acts on it.
Why we built it
Describing an app is often easier out loud than on a keyboard. Most people explain an idea more naturally by talking it through: "a booking page for a barbershop, three barbers, customers pick a time, I get an email." Typing the same thing tends to come out shorter and vaguer, and a vague prompt gets a vaguer first draft.
It also helps when typing is the slow part — on a laptop you're using one-handed, when your hands are busy, or when you simply think faster than you type. And it's useful while you look at the preview: you can describe what you see ("the button on the pricing card is cut off") without moving your eyes back and forth to type.
How to use it
- Open a project, or go to the home composer to start a new one.
- Click the microphone in the composer, or press Alt+V (Option+V on a Mac). The shortcut does the same thing as the button: it starts dictation, and pressing it again stops it.
- Allow microphone access the first time your browser asks. Without it, the browser can't hear you.
- Say one clear goal. For example: "Add a contact form to the about page with name, email and message fields."
- Stop dictation with the button or Alt+V again.
- Read the text and fix it. Speech recognition mishears names, technical words and punctuation. Edit the message box directly before you send.
- Send as usual.
Tips for good spoken prompts
Voice changes how you enter a prompt, not what makes a prompt good. The same advice applies:
- One goal per message. It's tempting to keep talking once you've started. A list of five changes in one breath is harder to act on — and harder to check afterwards — than five short messages.
- Say what should stay the same. "Change the colours, keep the layout" is clearer than "change the colours".
- Describe what you see. Point at a specific page or part of the preview rather than the app in general.
- Spell out the words that matter. Product names, brand names and unusual terms are the most likely to be misheard. Check them before you send, or type those parts.
- Edit, don't re-record. If one word is wrong, fixing it in the message box is quicker than dictating the whole thing again.
Our guide on how to write prompts for AI app builders goes further on phrasing.
What it doesn't change
Voice is a convenience for prompting, and nothing more:
- It doesn't change your mode. If you're in Build, Plan or Ask, a dictated prompt is handled exactly as a typed one would be. Pick the mode the same way you normally do.
- It doesn't change credits. A spoken prompt costs the same as the same prompt typed.
- It doesn't send on its own. You always see the text first.
Limits
- It depends on your browser's speech recognition. Voice uses the browser speech API, so how well it works — and whether it's available at all — comes down to the browser you use.
- It only appears where the shared prompt bar is used: the project workspace chat and the home composers built on the same control.
- Microphone permission is required. If you said no when the browser asked, dictation can't start. Re-enable the microphone for
mythex.aiin your browser's site settings and try again. - Transcription isn't perfect. Treat the text as a draft and read it before sending.
Get started
Next time you open a project, click the microphone, or press Alt+V, and describe the next change you want in a sentence or two. If the result isn't what you meant, you can always roll back with a checkpoint and try again.
The docs page is short and covers the details: Voice input.