VS Code is a fantastic editor, and for most of us it's where most of the workday happens. Typing comments, writing PR descriptions, hammering prompts into Copilot Chat — that's a lot of words, and a lot of micro-decisions about phrasing that quietly drain focus. Voice cuts that drain. The catch is that "dictation in VS Code" usually means juggling extensions, fighting whatever model is in fashion, or settling for the built-in voice, which is a step behind what your phone can do. Below is what actually works on a Mac in 2026, why I lean on an external tool over the built-in option, and how to wire it up in three minutes.

Why dictation is different in a code editor

Code editors are picky environments. You're not writing prose — you're managing cursor positions, dismissing auto-complete popups, jumping between tabs, pasting into chat panels. Most consumer voice tools assume the document is a Google Doc, and they get confused fast when the "document" is actually a comment block, a string literal, or a Copilot prompt. The things that matter for dictating in VS Code: - **Cursor awareness.** Text must land wherever the cursor is — top of file, inside a function, in the integrated terminal, in the chat sidebar. A system-wide hotkey is the only way to get this consistently. - **No extensions to babysit.** A "VS Code dictation extension" usually means another dependency, another account, another thing that breaks on the next VS Code update. An external tool bypasses the editor entirely. - **No weird character conversions.** "Open paren" needs to become `(`, not the word "open paren". A few tools do this well; most don't. - **Privacy.** Some devs are writing proprietary code. Audio going to Google's servers every time they say "fix the race condition in the webhook handler" is a non-starter. - **Offline.** Apple's built-in dictation needs internet for the more accurate mode. On a plane or a coffee shop with bad Wi-Fi, that's a problem.

Which is the best way to dictate in VS Code on Mac?

You basically have three options in 2026: 1. **VS Code's built-in voice typing.** Free, no install, but uses the Web Speech API through Chromium and is fairly basic. 2. **A VS Code speech extension.** Things like Microsoft's *VS Code Speech* extension or community forks. More featureful than the built-in, but you're married to the extension's update cycle. 3. **A system-wide dictation app** (like MetaWhisp) that sends text into the focused window. Works in VS Code, in the terminal, in your browser, in your chat app — anywhere your cursor happens to be.

How does VS Code's built-in voice typing work?

VS Code ships a built-in voice command called Voice: Start Dictation (and Voice: Stop Dictation) that you can trigger from the Command Palette. It's documented in the official VS Code voice guide. The honest version: - **Engine:** It uses the Web Speech API under the hood, which on macOS leans on Apple's dictation pipeline. That means audio can leave your Mac for the more accurate mode. - **Setup:** Open the Command Palette (⇧⌘P), type "Voice: Start Dictation", press Enter. Or assign a keybinding in Keyboard Shortcuts. - **Languages:** Whatever the underlying system supports — on macOS, that's the same set as Apple Dictation. - **Stop:** Press Escape, or use the "Voice: Stop Dictation" command. It's fine for a quick comment. It is not great if you want consistent punctuation, custom vocabulary, or a single button that behaves the same way in every app. That gap is what pushes most developers to option 3.

Should I use a VS Code speech extension instead?

If you want more control inside the editor, you can install a voice extension. The well-known one is Microsoft's **VS Code Speech** extension; there are also community alternatives. These add commands like voice.start to the palette and let you configure hotkeys and sometimes literal punctuation. Pros: - Native commands inside the Command Palette. - Some let you speak punctuation literally ("open paren" → `(`). - Works in any text file type VS Code opens. Cons: - Another extension to update, audit, and troubleshoot. - Some send audio to cloud STT APIs you can't always swap out. - You may have to enable per-workspace. If you live entirely inside VS Code and don't dictate in your browser, terminal, or chat tools, an extension is reasonable. Most developers — me included — dictate in too many places to live with that.

Method 3: A system-wide dictation app (the one I use)

A global dictation app doesn't care what window is focused. You press a hotkey, it captures your voice, it pastes the transcript into whatever has the cursor. In VS Code, that means it just works — in a comment, in a file, in the terminal panel, in the Copilot Chat sidebar. MetaWhisp is what I built for exactly this. It's a free macOS app that runs OpenAI's Whisper locally through WhisperKit on the Apple Neural Engine, with a single global hotkey (default Right Option ⌥). Why I built it that way: - **No account.** Install, run, dictate. Local mode is free and unlimited. - **Audio never leaves the Mac.** The Whisper model runs on-device. Your code comments aren't being shipped to a third party. - **One hotkey everywhere.** Same hotkey in VS Code, Chrome, Slack, BBEdit, anywhere. - **Free unlimited local mode.** No monthly minutes cap, no trial, no upsell on the core feature. - **Optional Pro.** If you want cloud accuracy on tough audio, or AI text polish via processing modes, Pro is $30/year or $7.77/month. Local mode stays free forever.

Founder's note: I run on an M1 Air with 8 GB of RAM. I have MetaWhisp set to launch at login with Right Option as the hotkey. I dictate 90% of my comments, PR descriptions, and prompts into Copilot Chat. The remaining 10% is actual code, where typing is faster than speaking.

How to set up MetaWhisp for VS Code on Mac in 3 minutes

You need a Mac running macOS 14 Sonoma or later on an Apple Silicon chip (M1, M2, M3, M4, or later). Intel is not supported because the model runs on the Neural Engine, which is Apple-only. **Step 1: Download and install.** Grab MetaWhisp from the download page — free, no signup. Open the .dmg, drag the app into Applications, launch it. macOS will ask for accessibility permission so the app can paste text into the focused window — click Allow. **Step 2: Download the Whisper model.** First launch triggers a one-time download of the Whisper large-v3-turbo model (~950 MB). It runs locally on the Neural Engine and never connects to a server during the download. The progress bar tells you when it's ready. **Step 3: Confirm your hotkey.** Default is **Right Option (⌥)**. Hold it, talk, release. To change it, open the app's settings — anything goes from F-keys to Caps Lock to a chord like Right Option + R. The global hotkey docs walk through it. **Step 4: Test it in VS Code.** Open VS Code, place your cursor in any file, hold Right Option and say: "remember to refactor this loop on Monday." Release. The text lands where the cursor was. Done.

How is this different from VS Code's built-in voice?

Feature	VS Code built-in	MetaWhisp local
Engine	Web Speech API / Apple Dictation	Whisper large-v3-turbo on ANE
Account needed	No	No
Internet required	Yes (for full accuracy)	No
Cost	Free	Free (local mode)
Audio leaves Mac	Yes (Apple servers)	No
Punctuation accuracy	Decent	Good, configurable
Works in terminal	Limited	Yes
Works in Copilot Chat	Limited	Yes
Languages	Apple-supported set	99 with auto-detect
Custom vocabulary	No	Pro feature

The big delta is offline + private. If you dictate comments about a customer system, an internal API, or a security-sensitive codebase, the audio going to Apple or Google every time is a real consideration. The general Mac dictation guide covers the privacy angle in more detail.

How do I dictate code comments in VS Code?

Code comments are where dictation pays for itself fastest. You think faster than you type when you can talk, and a good comment block reads better when it's natural language. The trick is to teach yourself a few voice-to-symbol replacements: - "open paren" → `(` - "close paren" → `)` - "open curly" → `{` - "equals" → `=` - "arrow" → `=>` - "colon colon" → `::` A typical multi-line comment workflow looks like this: 1. Type `//` and hit Enter to start the block. 2. Cursor is on line 2. Hold ⌥ and dictate: "this function handles user login, returns a token, throws on invalid credentials." 3. Release. Edit a couple of words. Move on. The transcript lands at high accuracy on clear English in my own testing, so editing is usually a quick pass, not a full re-type. If you dictate a lot of code, Pro's Structured processing modes can strip filler words, fix casing, and reformat dictated code blocks.

Pro tip: If you have to write a long function comment, dictate it in two passes: first the summary sentence, then a second pass for the edge cases. Splitting it up reduces mid-thought filler words and gives you cleaner output to edit.

Can I dictate Copilot Chat prompts?

VS Code's Copilot Chat sidebar is a normal text area. Your cursor is there, you press a hotkey, you talk — the prompt goes in. For prompts longer than a sentence this is a huge quality-of-life win, because long prompts are where you tend to typo, drop words, and lose your train of thought. A worked example: 1. Click into the Copilot Chat input box. 2. Hold Right Option and say: "given this React component, refactor it to use the new server actions API, and explain the security tradeoffs of using form actions versus client side fetch." 3. Release. The prompt appears, you hit Enter, Copilot does its thing. Same workflow in Cursor, in the integrated terminal, in a pull request description on GitHub.com, in Slack — anywhere with a text field.

Why isn't my VS Code dictation working?

A handful of things go wrong on day one. Here are the fixes I've hit personally and watched other people hit. **"The hotkey fires but nothing pastes."** macOS is blocking the paste. Go to System Settings → Privacy & Security → Accessibility and make sure MetaWhisp is on the list. Toggle it off and back on if it was already enabled — that fixes a lot of stale-permission issues. **"It's slow on my M1."** The first transcription is the slowest because the model has to warm up. After that it's typically faster than realtime on an M1 Pro or later. If it's slow on a base M1, switch the language in settings to your actual language instead of auto-detect — that saves a few hundred milliseconds. **"It keeps inserting 'period new line' when I want to keep talking."** The hotkey is the boundary. If you want a longer dictation, switch to "press to start, press to stop" mode in settings. **"VS Code keeps stealing focus and text goes to the wrong tab."** This is usually a macOS focus issue, not a MetaWhisp issue. In System Settings → Desktop & Dock, turn off "Focus follows mouse" if it's on, and make sure VS Code is the active window before you press the hotkey. **"I need to dictate a different language in the same session."** MetaWhisp supports 99 languages with auto-detect. If you want a fixed language for a session, set it in settings. The Pro plan adds translation.

How to choose for yourself

If you only dictate occasionally, the built-in VS Code voice is a fine starting point. If you dictate every day and care about privacy, offline operation, or accuracy on technical vocabulary, a system-wide tool like MetaWhisp is worth the three minutes. The free local mode covers what most developers need; the Pro plan exists for cloud accuracy on tough audio, AI text polish, and translation. Either way, the upgrade in flow state is real. Once you can talk to your editor, you stop optimizing keystrokes and start thinking about the actual problem.

FAQ