Vibe Prompt

Private, local speech-to-text for prompts and code in VS Code. Vibe Prompt records from the local microphone, shows rolling transcript previews in a bottom panel, and copies the final transcript to the clipboard when recording stops.

Supported Platforms

The production target is VS Code Desktop on Linux, Windows, and macOS:

Vibe Prompt is a local UI extension. Microphone capture and transcription run on the machine hosting the VS Code interface, including when the workspace is opened through WSL, Remote SSH, or a Dev Container. The local VS Code host, not the remote workspace, determines which runtime package and audio capture implementation are used.

Requirements

• VS Code 1.90 or newer
• FFmpeg available on PATH
• Linux, Windows, or macOS on x64 or ARM64

Audio and transcription stay on the local machine. The extension does not read workspace files or send audio to an external service.

Setup

1. Install FFmpeg. Vibe Prompt verifies the executable and required audio backend during setup. If FFmpeg is not on PATH, configure Vibe Prompt: FFmpeg Path.
2. Open the Vibe Prompt panel and install the local whisper.cpp runtime. Linux and Windows use checksum-verified official archives. macOS shallow-clones whisper.cpp's latest default branch and builds it locally; this requires Git, CMake, and Xcode Command Line Tools.
3. Choose Tiny English, Base English, or Small English. The extension downloads the model to its private storage, verifies its size and SHA-256, and configures it automatically.

Platform prerequisites:

• macOS: install Homebrew, then run brew install cmake ffmpeg and xcode-select --install.
• Windows: install an FFmpeg build with DirectShow support, then use Vibe Prompt: FFmpeg Path if it is not on PATH.
• Linux: install FFmpeg through the distribution package manager and ensure PulseAudio input support is available.

Base English is the recommended default. Model Path remains available in advanced settings for an existing or custom whisper.cpp model.

On Ubuntu, the default audio input uses PulseAudio/PipeWire through FFmpeg's default input. Change Vibe Prompt: Audio Input if the microphone has a different source name.

On Windows, default discovers DirectShow microphones and initially uses the first available input. Use the microphone button in the transcript toolbar or run Vibe Prompt: Select Microphone to choose a different device.

On macOS, default discovers AVFoundation microphones and initially uses the first audio input. Use the microphone button to select another input. macOS may ask for microphone access the first time FFmpeg records; allow access for VS Code and restart recording.

Use

Open the Vibe Prompt panel from the bottom panel area and select Start. Partial transcript text appears while recording. Select Stop to run a final transcription and copy it to the clipboard, then paste it into your editor or AI chat composer.

The panel reports each session phase explicitly (starting, recording, stopping, transcribing, or cancelling) and ignores conflicting actions while a transition is in progress. Live text is provisional and may revise itself as more speech provides context. To keep local CPU use bounded, live preview pauses after one minute; stopping always transcribes the complete recording.

The completed transcript remains editable in the panel. Copy always uses the latest edited text.

Use the model selector in the transcript toolbar to switch between installed models. Open Models to download another model or remove an installed model that is not currently active.

If an active model is removed outside VS Code, the panel detects it, reports the missing file, and returns to model setup without starting a recording.

The default start/stop shortcut is Ctrl+Alt+Space. Press Escape while recording to cancel.

Direct insertion into third-party chat composers is not used because VS Code extensions do not have a common supported API for their prompt inputs.

Development

npm install
npm run check
npm run compile
npm test
npm run package
npm run package:linux-x64
npm run package:linux-arm64
npm run package:win32-x64
npm run package:win32-arm64
npm run package:darwin-x64
npm run package:darwin-arm64

The platform-targeted commands produce Marketplace-compatible VSIX packages for the supported desktop architectures.