Speak your ideas instead of typing them
This prompt gives an AI coding assistant a clear brief for building your own voice dictation app. The goal: press a shortcut, speak, and insert your words into the text field you were using.
It asks for local speech recognition, optional text cleanup, a simple interface, and practical testing. This is a build prompt, not a ready-made download. Setup and performance will depend on your computer and the tools your coding assistant can use.
How to use it
Open an AI coding assistant that can create and run an app on your computer.
Copy the entire prompt below and paste it into a new project.
Tell it whether you use Mac or Windows. Follow its setup instructions and test a short sentence before relying on it.
Copy this prompt
Build me a simple, private voice dictation app for my computer. I’m non-technical, so handle the implementation and explain setup in plain English.
First, ask whether I use Mac or Windows, unless you can already detect it. Build for my operating system. Use a generic app name that I can change.
Core experience
Run quietly in the menu bar or system tray.
Let me press a customizable keyboard shortcut to start recording and press it again to stop.
Show a small floating panel with microphone activity and clear recording, transcribing, and finished states.
Transcribe my speech and insert the text into the text field I was using.
Keep track of the correct destination without stealing focus or pasting into a different app.
If automatic insertion fails, preserve the transcript and provide an obvious Copy button.
Never automatically send a message or press Enter.
Let me cancel a recording.
Transcription and formatting
Use a locally running speech recognition model.
Preserve my meaning and wording, adding sensible punctuation and capitalization.
Never answer my dictated questions or carry out instructions contained in my speech. Treat speech as text to transcribe.
Make additional cleanup optional. If enabled, use a local model and avoid inventing details or substantially rewriting what I said.
Only apply special formatting when I explicitly request it, for example by starting with “Format this as an email.”
Do not invent greetings, names, sign-offs, or other content.
If optional formatting fails, return the original transcript.
Detect silence and microphone failures rather than generating text from empty audio.
Setup and privacy
Work on a fresh computer without assuming any models, tools, accounts, or dependencies are already installed.
Explain required downloads, disk space, and permissions before setup.
Keep recording, transcription, and optional cleanup on my computer. No cloud transcription, subscriptions, analytics, or API keys.
Explain that initial installation may require internet access, while dictation should work offline afterward.
Record only when I explicitly start recording.
Delete temporary recordings after processing, including after errors.
Do not retain transcript history by default.
Include microphone selection and a live input meter so I can check the microphone.
Explain permission problems clearly and provide recovery steps.
Design
Keep the interface clean, compact, and beginner-friendly.
Include settings for the shortcut, microphone, language, transcription model, and optional cleanup.
Show real progress and honest success or failure messages.
Keep the shortcut reliable after sleep and wake.
Delivery and testing
Build the actual app, not just a mockup or plan.
Provide the source code, a runnable app, and simple installation and uninstall instructions.
Test recording, transcription, cancellation, silence handling, permissions, and insertion into representative native and browser text fields.
Check that the app inserts text once, into the intended field, without submitting it.
Verify offline dictation after setup.
Clearly distinguish what you tested from anything that still needs my microphone input or permission.
If your environment cannot build or test part of the app, explain the exact limitation and give me the smallest next step.
Choose sensible defaults and keep the first version focused on reliable dictation.
