VS Code Local Dictation: What Developers Should Test

VS Code now offers local dictation. Test where it works, where it stops, and whether a system-wide voice workflow fits better.

Recent coverage has brought attention to a quiet VS Code feature: built-in dictation powered by a model running on your local device. It’s worth trying if you already spend most of your day in the editor. The better question isn’t whether voice can fill a Copilot prompt. It’s when voice stops being useful.

XDA covered the feature in a September 13 report. Microsoft says VS Code downloads a speech-recognition model to your device the first time you use it. After that, it can transcribe speech locally, even without an internet connection. You can use it in the editor, terminal, chat input, source-control commit box, and pull-request comments. [1] [2]

What arrived in VS Code

The local model keeps one small part of the workflow off the network. That helps on a train, in a remote workspace, or whenever you don’t want to send every rough prompt to another service. Microsoft says dictation runs on the local VS Code client, even with remote workspaces. [2]

These are two separate features. Dictation turns your speech into editable text. Voice Mode lets you speak with an agent, which reads its replies aloud and may use the active chat session and selected model. They’re easy to confuse because both use a microphone button, but they do different things.

For plain dictation, VS Code offers a handy push-to-talk option. Hold the shortcut while you speak, then release it when you're done. Use Ctrl+Alt+V to start dictation in the editor. In chat, press Ctrl+I to start or stop dictation. Terminal dictation also cleans up commands for the command line.

The first test should be boring

Don’t judge a voice feature by asking it to build something from a vague sentence. Test the small tasks that cause friction every day.

Start with a short bug report that includes a file name, a function name, one command, and a correction. Review the transcript before submitting it. Then dictate a commit message and a reply to the pull request. Finally, try a terminal prompt with a flag, a path, and punctuation you would normally type.

That test makes the tradeoff clear. A local model can be fast and private in one limited way because speech recognition runs on your device. But that doesn't mean the rest of the process stays there. Sending a dictated prompt to an AI chat session is a separate choice. Developers should explain that boundary clearly instead of treating a microphone icon as proof that everything is private.

The editor is not the whole workday

This is where a VS Code-only feature starts to show its limits. Coding rarely happens in one window. In a typical hour, you might move between the editor, a browser ticket, Slack, a code review, a terminal, an issue tracker, and an email asking someone to test a build. The built-in feature can make work in the editor easier, but it can't follow you elsewhere.

That’s not a criticism of VS Code. It’s a reasonable limit for an editor feature. It also helps you decide whether voice input should be an input layer, not just another editor command. If you only need it for writing code or Copilot prompts, try the built-in tool first. It’s free and already in the editor.

If the same problem occurs in a browser, Slack, Outlook, or a remote desktop, try a system-wide dictation tool. DictaFlow is built for this. Hold a hotkey, speak, and release it. The text appears wherever the cursor is active, in any app. Its Windows typing mode can help in Citrix, VMware Horizon, and RDP sessions, where clipboard insertion often fails. This compares workflows. It does not suggest that an editor should become a desktop dictation app.

A sensible setup path

Start with a simple setup. Turn on dictation, choose the right microphone, and test it in the editor before adjusting cleanup settings. Microsoft documents an optional language model that can fix punctuation, capitalization, paragraphs, lists, and number formatting. Leave it off at first if you want to see the raw speech recognition output. [2]

Then see which terms need extra help. VS Code lets you add dictation instructions to a user or workspace file. This can help with projects that use product names, APIs, and common abbreviations. Test the words you find hardest to say. A smooth demo sentence won’t show how the tool handles the names that slow you down.

One detail is easy to miss: the on-device model doesn’t run on every platform. Microsoft currently supports Windows x64 and Arm64, Apple silicon Macs, and supported Linux x64 and Arm64 systems. It doesn’t support VS Code for the Web, Intel Macs, 32-bit or Arm32 systems, or musl-based Linux distributions such as Alpine. Microsoft recommends the VS Code Speech extension for those platforms. [2]

The practical takeaway

The XDA report points out a feature many developers may have missed. Local dictation in a serious editor is no longer just an experiment. It could save you a few minutes of typing in your next coding session.

Don’t make it sound like more than it is. Try VS Code dictation for editor text, terminals, commits, and prompts. Keep the transcript visible, and check technical names before sending. If you want the same hold-to-talk habit in other text fields, test a system-wide tool too. The best setup cuts down on retyping without creating a new mess to clean up.