VS Code Voice Assistant

Summary

What it is

VS Code Voice Assistant brings a ChatGPT-like microphone workflow into VS Code. The current slice records speech from the default Windows microphone, transcribes it locally with whisper.cpp, accumulates the raw transcript in a Secondary Side Bar view, and keeps audio on the user's machine.

Why it exists

The problem behind it

This project exists because speaking to ChatGPT in the browser is useful, but that workflow breaks when I am already deep inside VS Code with coding agents and editor context. I wanted a voice workflow that feels native to the editor, works locally, and eventually turns messy spoken intent into clean prompts or code-related text.

Brief architecture

How it is put together

The extension contributes a Secondary Side Bar webview, status-bar microphone shortcut, and recording commands. On Windows, it captures PCM audio from the default microphone through a native WinMM helper, writes a temporary WAV file, runs local whisper.cpp transcription, appends the raw transcript to the webview state, and deletes the temporary audio file after processing. Runtime and model downloads are pinned and checksum verified.

Highlights

What this project covers

• Adds a compact Voice Assistant view in VS Code's Secondary Side Bar.
• Captures native PCM microphone audio on Windows and transcribes locally.
• Automatically provisions the pinned whisper.cpp runtime and base English model with checksum verification.
• Avoids backend accounts, API keys, telemetry, and audio uploads.