How It Works
Using SpeakUp with Cursor takes three steps:
- Click into Cursor's chat panel (or any text field — inline edit, terminal, command palette).
- Press your SpeakUp hotkey and speak your prompt. Describe what you want Cursor to build, fix, or refactor.
- Press the hotkey again. Your words appear as typed text in the chat panel. Hit Enter, and Cursor generates the code.
SpeakUp simulates keystrokes directly — no clipboard, no paste. It works in every Cursor text field, including the chat panel, inline edit prompts, the integrated terminal, and the command palette.
Why Not Cursor's Built-In Voice
Cursor introduced an experimental voice input feature, but the developer community has found it unreliable. Forum posts and community discussions report that 70 to 80 percent of voice submissions arrive empty — no transcription at all. The feature requires a network connection and sends audio externally for processing.
SpeakUp uses OpenAI's Whisper model running locally on your Mac's GPU via whisper.cpp and Metal acceleration. There is no network dependency. Transcription is consistent and accurate, whether you are on a plane, behind a corporate firewall, or working from a cafe with poor WiFi.
Why Not Wispr Flow
Wispr Flow markets itself to developers, but it introduces two problems that matter when you are writing code:
Screen capture. Wispr Flow takes screenshots of your active window and sends them to cloud servers. When you are working in Cursor, that means your code, your file tree, your terminal output, and anything else visible on your screen is transmitted externally. If you work on proprietary code, client projects, or anything under NDA, this is a security risk.
Prompt rewriting. Wispr's AI "auto-edits" your dictation, rewriting your words before inserting them. For a Cursor prompt, precision matters. If you say "add a try-catch around the Stripe webhook handler," you need exactly those words — not a paraphrased version that loses the specific function name or service reference.
SpeakUp has zero screen access, zero network calls, and zero AI rewriting. Your words go in exactly as you speak them.
The SpeakUp Advantage
On-device processing. Audio is processed by whisper.cpp on your Mac's GPU. Nothing leaves your machine. Your code, your prompts, and your voice data stay private.
Faithful transcription. SpeakUp transcribes exactly what you say. No paraphrasing, no auto-correction, no AI "improvements." Technical terms, framework names, and specific instructions come through accurately.
One price, forever. SpeakUp costs €29 once. No subscription, no monthly billing. You already pay $20 per month for Cursor Pro — you do not need another recurring charge for voice input.
Prompts That Work Well with Voice
Dictation is especially effective for detailed, descriptive prompts — the kind that produce the best results from Cursor but feel tedious to type:
- "Add pagination to the user list component with 20 items per page and show the total count in the footer"
- "Write a unit test for the calculateTax function that covers edge cases for negative values and zero input"
- "Refactor this component to use React Query instead of the manual useEffect fetch pattern"
- "Create a middleware that validates the JWT token and returns a 401 with a clear error message if it is expired"
- "Add a loading skeleton to the dashboard that matches the layout of the actual content"
Each of these prompts takes 5 to 10 seconds to speak. Typing them takes 30 to 60 seconds. Over a full day of vibe coding, the time saved is substantial.
Related: Voice Dictation for Vibe Coders · SpeakUp for Developers