VoicePrompt Dev
Speak prompts directly into VS Code Chat, Codex, your terminal, or the active editor.
VoicePrompt Dev records audio locally, transcribes it with OpenAI or your own command, and sends the result where you are already working. It is built for the fast loop of thinking, speaking, and coding without leaving VS Code.
Quick start
- Install VoicePrompt Dev from the VS Code Marketplace.
- Open Command Palette and run
VoicePrompt Dev: Open.
- Set an OpenAI API key in your User Settings (or configure your own transcription command), then record with
R or hold Space.
- Choose whether the result should go to your active terminal, chat tab, or editor.
The default transcription backend is OpenAI. A ChatGPT subscription does not include OpenAI API usage; create an API key and add billing or credits to your OpenAI Platform account.
{
"voicePromptDev.openaiApiKey": "your-openai-api-key"
}
Keep this key in user settings only—never commit it or place it in workspace settings. If you prefer a private or local transcription pipeline, use the command backend instead.
What it does
- Records locally on Windows and Linux, with browser-microphone fallback for macOS and other environments
- Transcribes with the OpenAI audio API or a custom command
- Routes transcripts to the active VS Code terminal, chat tab (including Codex when active), or editor
- Supports tap-to-record, hold-to-record, replace/append, and optional auto-submit
- Keeps common terminal approval actions available while the panel remains focused
- Uses a compact VS Code-native panel that adapts to light and dark themes
Built for agent workflows
Use it to dictate a task into Codex or another chat provider, send a command to a terminal, or turn a half-formed thought into the next line of an editor. The panel can keep your selected destination in focus and avoids forcing you through a separate browser workflow.
Privacy and security
- Audio is recorded locally and temporary capture files are removed after transcription or cancellation.
- Audio is sent to OpenAI only when the OpenAI backend is selected.
- A custom transcription command is available for local or private pipelines.
- Sensitive settings are machine-scoped so workspace settings cannot inject transcription commands or chat command IDs.
- API keys are not passed on a subprocess command line.
Shortcuts
| Shortcut |
Action |
R |
Start or stop recording |
Hold Space |
Push to talk |
T |
Focus the selected terminal or chat target |
Ctrl+Shift+V |
Restore focus to VoicePrompt |
E |
Send Enter to the active terminal |
Y, N, 1, 2, 3 |
Send approval keys to the active terminal |
Configuration
The useful settings are:
voicePromptDev.transcriptionBackend: openai (default) or commandvoicePromptDev.openaiApiKey: your OpenAI API keyvoicePromptDev.openaiModel: transcription model, default gpt-4o-mini-transcribevoicePromptDev.openaiLanguage: optional language hint, for example envoicePromptDev.transcriptionCommand: command that receives {input} and prints only the transcriptvoicePromptDev.recordingMode: auto, native, or browser
For a custom transcription backend:
{
"voicePromptDev.transcriptionBackend": "command",
"voicePromptDev.transcriptionCommand": "my-transcriber --input {input}"
}
- Windows: bundled recorder with a PowerShell fallback. If a recording is silent, confirm the selected input device in Settings > System > Sound > Input.
- Linux: uses
parec or arecord when available.
- macOS / fallback environments: uses browser microphone capture inside the panel.
For detailed recovery steps and microphone-permission help, see TROUBLESHOOTING.md. Known limitations and development notes are available in the GitHub repository.
Feedback
If VoicePrompt Dev makes your coding workflow smoother, a Marketplace rating or review helps other developers find it. Bug reports and feature requests are welcome in the issue tracker.
License
MIT. See LICENSE.
