CodeWhisper (VS Code)

In the Extensions sidebar and Marketplace, this extension is shown as CodeWhisper (displayName in package.json). The shorter field name is only the technical id (codewhisperer.codewhisper-voice); it does not replace that title.

Ask questions by voice about the code in your active editor. The extension sends your speech transcript plus the current file or selection to Groq, then plays the answer with ElevenLabs text-to-speech.

Not affiliated with Amazon Web Services or AWS CodeWhisperer. This is an independent open-source extension.

Requirements

• Visual Studio Code 1.85+
• Python 3 on your PATH (python3 on macOS/Linux, python on Windows unless you override it)
• Python package httpx (see Python setup)
• Groq API key and ElevenLabs API key + voice ID

Python setup

The extension runs a small backend from the installed extension folder:

python3 -m pip install --user -r /path/to/extension/backend/requirements.txt

Easiest from a clone of this repo:

cd codewhisper/backend && python3 -m pip install --user -r requirements.txt

For local development you can instead use a venv inside codewhisper and point CodeWhisper › Python Path at codewhisper/.venv/bin/python (macOS/Linux) or .venv\\Scripts\\python.exe (Windows).

After installing the extension from the Marketplace or a .vsix, use either:

• Command Palette → CodeWhisper: Install Python Dependencies (runs pip install --user against backend/requirements.txt), or
• When the backend fails, choose Install now or Copy pip command from the prompt.

If you use a non-default interpreter, set CodeWhisper › Python Path in Settings.

Extension settings

Usage

1. Configure the three API-related settings above.
2. Open a file and focus the editor.
3. Run CodeWhisper: Voice Ask from the Command Palette, click the CodeWhisper status bar item, or use Cmd+Shift+Space (macOS) / Ctrl+Shift+Space (Windows/Linux) when the editor is focused.
4. In the side panel, either tap the microphone and speak (Web Speech API), or type a question and click Ask with text (same pipeline, no microphone).

If nothing is selected, the whole file is sent as context (truncated on the server side). With a selection, only the selection is sent.

Speech recognition limitations

Recognition runs in an embedded webview (Chromium). Availability and quality depend on OS, VS Code version, and network (where the engine requires it). If you see “Speech API not supported”, use a different environment or keyboard input (future versions may add alternatives).

“Mic error: not-allowed”

That means the OS denied the microphone to the editor process (not your Groq/Eleven keys).

• macOS: System Settings → Privacy & Security → Microphone — enable Visual Studio Code and/or Cursor (whichever you use). If it was off, turn it on, then in VS Code run Developer: Reload Window and try Voice Ask again.
• Windows: Settings → Privacy & security → Microphone — allow microphone access and allow Visual Studio Code for desktop apps.
• Linux: varies by distro; allow the browser/Electron app to use the mic (PipeWire/PulseAudio, Flatpak portal, etc.).

If you previously clicked Don’t allow, reset the choice in the same privacy screen or remove/re-add the app from the list where applicable.

macOS never shows VS Code / Cursor under Microphone

• The list often only fills after the app has tried to use the mic. Try Voice Ask → mic once, then reopen System Settings → Privacy & Security → Microphone and scroll the full list (VS Code and Cursor are separate entries).
• Cursor and Visual Studio Code are different apps — enable the one you actually launch.
• Some Electron builds still never trigger the system prompt from a webview; use Ask with text in the CodeWhisper panel to verify Groq + ElevenLabs anyway.
• To reset only VS Code’s mic decision (Terminal):
  tccutil reset Microphone com.microsoft.VSCode
  Then restart VS Code and try the mic again (you may need to approve once). For Cursor, get its bundle id with osascript -e 'id of app "Cursor"' and run tccutil reset Microphone <that-id> the same way.

ElevenLabs 401 Unauthorized

That means ElevenLabs rejected your API key (wrong, expired, or copied with extra characters).

1. In ElevenLabs, open API keys (from your profile / developer settings) and copy the key used for HTTP requests (xi-api-key).
2. In VS Code: Settings → Codewhisper: Elevenlabs Api Key — paste only the key (no Bearer prefix, no quotes). The extension trims leading/trailing spaces (v1.0.2+).
3. Developer: Reload Window and try Ask with text again. The voice ID must belong to the same ElevenLabs account as that key.

Privacy

• Transcript and code context are sent to Groq for the model response.
• Model output text is sent to ElevenLabs to synthesize audio.
• Do not use with code or speech you are not allowed to send to those services.

Logs from the Python process appear under Output → CodeWhisper.

Development

cd codewhisper && npm install && npm run compile

Press F5 (Run Extension) from the codewhisper folder.

Publishing

Source: github.com/pulkitgovrani/codewhisper. The publisher field in package.json must match the publisher id shown in Visual Studio Marketplace or Open VSX (it is currently set to codewhisperer; if vsce rejects it, use the exact id from your publisher page).

From the codewhisper folder, use a clean out/ so old builds are not bundled:

rm -rf out && npm install && npm run compile && npx @vscode/vsce package

This produces codewhisper-voice-<version>.vsix (from the name and version fields in package.json).

Mic troubleshooting: the panel first calls getUserMedia so macOS can show a real microphone prompt; only then it starts Web Speech. If you still see not-allowed, use Ask with text or enable the app under System Settings → Privacy & Security → Microphone and Developer: Reload Window.