VS Code Dictation in 2026: Built-In Voice Setup Guide

VS Code dictation changed in August 2026. Version 1.132 now supports multilingual, on-device speech recognition in chat inputs, writable editors, and the integrated terminal. On supported desktop systems, you no longer need the VS Code Speech extension for basic dictation.

The update makes voice part of the editor instead of sending you to a separate transcription window. You can speak a Copilot prompt, add prose to a Markdown file, or dictate a terminal command without switching apps. Microsoft also added shell-aware cleanup. For example, "git commit dash m hello world" can turn into a usable command instead of a literal transcript.

This guide explains how to set up dictation in VS Code, its hardware limits, and when a system-wide tool makes more sense.

What changed in VS Code 1.132

Microsoft released VS Code 1.132 on August 5, 2026. According to the official release notes, built-in dictation now uses the multilingual Nemotron 3.5 model on the local device. It follows the language set in agents.voice.language, or uses the system locale and automatic detection when supported.

The first-use screen now shows a live microphone waveform and a device picker. This helps when a developer machine has a laptop microphone, headset, webcam, and audio interface connected at the same time. You can reopen the introduction from the Command Palette with Voice Mode: Show Introduction.

VS Code can also load dictation instructions from ~/.copilot/dictation.md and .github/dictation.md in trusted workspaces. Teams can use these files to add project terminology and formatting preferences without changing the speech model.

How to turn on VS Code dictation

Built-in dictation is experimental and requires AI features to be enabled. The dictation.enabled setting is on by default, but an organization can manage it. If the microphone controls are missing on a managed computer, check this setting before reinstalling anything.

In Chat or the Agents window, select the microphone in the input box. You can also press Command+I on macOS or Ctrl+I on Windows and Linux. Press the shortcut again to stop recording. VS Code inserts the transcript without submitting the prompt, so you can review commands, file names, and numbers first.

In a writable editor, run Voice: Start Dictation in Editor from the Command Palette. The default shortcut is Option+Command+V on macOS, or Ctrl+Alt+V on Windows and Linux. A microphone control appears near the cursor during recording.

The first time you use it, VS Code downloads the speech model to your device. After the download finishes, speech recognition works offline. If a corporate network blocks the download, VS Code 1.132 can import an official Foundry Local model package from disk.

Check platform support before troubleshooting

The on-device model supports Windows on x64 and Arm64, Apple silicon Macs, and Linux x64 or Arm64 systems with glibc 2.34 or later. It also works with remote workspaces because speech recognition runs in the local VS Code client.

The built-in model doesn't support VS Code for the Web, Intel Macs, or musl-based Linux distributions such as Alpine. Microsoft recommends the VS Code Speech extension for unsupported platforms. The extension also provides features missing from the basic built-in path, including voice chat, spoken chat responses, and the "Hey Code" activation phrase.

Don't confuse a remote workspace with a remote desktop. A local VS Code window connected to a remote repository can still use your local microphone. A full RDP, Citrix, or VDI session may apply separate microphone and clipboard policies. Test that environment directly.

Add project vocabulary without fighting every correction

Developer dictation gets names wrong more often than regular prose. Repository and package names, acronyms, camelCase identifiers, and internal product terms can turn a quick prompt into a correction job.

Run Voice: Configure Dictation Instructions to create or edit instruction files. Put shared terms in .github/dictation.md so the repository keeps them. Store personal preferences in ~/.copilot/dictation.md. Keep the list short instead of building a huge glossary. Start with the ten terms you say every day, then add only errors that keep recurring.

Test the vocabulary in three places: Copilot Chat, a Markdown editor, and the terminal. The same spoken phrase may need different output in each one. Terminal cleanup must preserve shell syntax, while a document needs normal punctuation and capitalization.

Where built-in VS Code dictation stops

The new feature is free, runs locally, and fits well with VS Code. For developers who do all their work in the editor, that may be enough.

The workday doesn't end at the editor. You may still need to explain a bug in Jira, respond to a GitHub review, write a Slack update, send an email, or add notes in a browser-based admin tool. VS Code dictation won't follow your cursor into those apps.

The difference is where the text needs to go. Editor dictation works best when you already know the destination. A system-wide input tool works better when you switch destinations every few minutes.

When DictaFlow fits the developer workflow

DictaFlow adds to VS Code rather than replacing its built-in feature. It works on Mac, Windows, and iOS, and you can use it on Android through Telegram. Hold a shortcut, speak, then release it. Your text appears at the active cursor in the app you're using.

App-aware formatting adjusts cleanup for the destination. A coding prompt, support ticket, email, and clinical field each need different formatting. The Knowledge Base and custom vocabulary also help handle package names, internal tools, client names, and technical phrases.

DictaFlow Pro costs $7 per month or $69 per year. Its Local Offline mode works without an internet connection once you download the model. Cloud processing stays private and offers the best overall quality with the full feature set. In Citrix, RDP, VDI, and other clipboard-hostile fields, DictaFlow can simulate keystrokes instead of relying on paste.

DictaFlow comparison guide covers the main alternatives. The getting started guide explains how to set up hold-to-talk and test text insertion in your usual apps.

Run one practical voice coding test

Don't judge VS Code dictation by one clean sentence. Test it for ten minutes with the kind of work you want it to handle.

First, enter a Copilot prompt that includes a file path, two package names, and an exact version number. Then add a short Markdown explanation with punctuation and a numbered list. Next, enter a safe terminal command and review it before running anything. Finally, repeat the same task in Jira, GitHub, Slack, or another app outside VS Code.

Measure how long corrections take, not just recognition accuracy. Check whether the microphone starts reliably, the correct device remains selected, and text appears at the intended cursor. If VS Code handles every destination you need, use its built-in tool. If problems begin outside the editor, test a system-wide option with the same script.

The practical choice for developers

VS Code 1.132 makes built-in dictation a practical option for developers. Its on-device model, multilingual support, project instructions, and shell-aware terminal cleanup are major upgrades. Start there because it's already part of the editor and costs nothing.

Use DictaFlow when voice input must keep working after you switch to a browser, email, chat, remote desktop, or another machine. The best choice depends less on raw transcription speed than on how many places you actually work.