1. Four Copilot surfaces, four different voice stories
Copilot’s chat and agent features have a separate text input in every host: the Chat view and Agents window in VS Code, the Copilot tool window in IntelliJ IDEA, PyCharm, WebStorm, Rider and the other IntelliJ-based IDEs, the chat window in Visual Studio, and the prompt in Copilot CLI. Voice support belongs to the host, not to your Copilot subscription, so a feature you use in one editor may simply not exist in another.
Before installing anything, find your row below. If the host already has dictation, use it; a separate app is only worth adding where there is a gap or where you want the same key everywhere.
- VS Code desktop: built-in dictation in chat, the Agents window, editors and the terminal (experimental), plus an experimental Voice Mode.
- VS Code for the Web: dictation streams audio to the Microsoft AI voice service and needs a network connection and GitHub sign-in.
- Copilot CLI: /voice downloads a speech model and transcribes on your machine; it has its own guide.
- JetBrains IDEs: no microphone in the Copilot plugin; voice-input requests are still open on Microsoft’s feedback tracker.
- Visual Studio: Microsoft’s Copilot Chat article (last updated May 14, 2026) describes no dictation control.
2. VS Code: speak a Copilot Chat prompt with the on-device model
Microsoft’s voice documentation, updated September 30, 2026, says built-in dictation is available when AI features are enabled and is on by default through the dictation.enabled setting. The first time you use it on the desktop, VS Code downloads a speech model named nemotron-3.5-asr-streaming-0.6b; after that, recognition runs locally and offline. The model is offered on Windows (x64 and Arm64), macOS on Apple silicon, and Linux on x64 and Arm64 with glibc 2.34 or later. It also works in remote workspaces, because recognition happens in the local VS Code client rather than on the remote host.
In the chat input, select the microphone or press ⌘I (Ctrl+I on Windows and Linux), speak, and press the same shortcut to stop. The words go into the input without submitting it, and Escape throws away the text from that dictation session. For an editor, run Voice: Start Dictation in Editor or press ⌥⌘V (Ctrl+Alt+V); the same works in the Source Control commit box and pull request comment fields, and holding either shortcut turns it into push-to-talk. A single session stops on its own at 20 minutes and keeps what it has transcribed.
If you run Copilot CLI in VS Code’s integrated terminal, run Voice: Start Dictation in Terminal from the Command Palette. Microsoft says terminal dictation adapts speech for the command line: it drops ordinary punctuation, converts spoken symbol names and adjusts the first letter’s capitalization.
- Chat or Agents window: ⌘I on a Mac, Ctrl+I on Windows and Linux.
- Editor, commit message or PR comment: ⌥⌘V on a Mac, Ctrl+Alt+V elsewhere.
- Terminal: Voice: Start Dictation in Terminal from the Command Palette.
- Pick the microphone with Chat: Dictate: Select Microphone; it applies to every dictation surface.
3. Check where the transcript goes, not only the audio
On the desktop the audio stays on your computer, but the text may not. The setting dictation.experimental.llmCleanup is on by default, and while it is on, VS Code separately sends the finished transcript to a Copilot language model to fix punctuation, capitalization, paragraphs, lists and number formatting. Turn it off if you want the transcript to stay local as well. Administrators can enforce both choices with the DictationModel and DictationLLMCleanup policies; requiring the on-device model makes dictation unavailable in VS Code for the Web.
While cleanup is on, you can steer it with plain-language instructions: run Voice: Configure Dictation Instructions to create ~/.copilot/dictation.md for all workspaces, or .github/dictation.md for one trusted workspace. That is the place to list preferred spellings of your product and library names. On a network that blocks the model download, Chat: Install Dictation Model from Local Package installs the official CPU model from a ZIP file.
Built-in dictation is not offered on Intel Macs, 32-bit and Arm32 systems, or musl-based Linux such as Alpine. There, Microsoft points to the VS Code Speech extension, which also provides voice chat, read-aloud chat responses and the “Hey Code” keyword activation. Its Marketplace page says no internet connection is required and audio is processed locally, lists 26 languages, and on Linux supports Ubuntu 20.04, 22.04 and 24.04, Debian 11 and 12, RHEL 8 and CentOS 8 with the libasound library.
4. VS Code Voice Mode: talk with the agent, within plan limits
Voice Mode is a different feature from dictation. Instead of only turning speech into text, it reads the agent’s replies aloud and listens for your follow-up. Enable agents.voice.enabled, then select the Voice Mode button in the chat input or press ⇧⌘Space (Ctrl+Shift+Space on Windows and Linux). Requests go to the active chat or agent session with its selected model and attached files, and Voice Mode can also start a new session or answer questions about running ones. With agents.voice.handsFree on, it starts listening again once it finishes speaking; speak or press the shortcut to interrupt, and use ⇧⌘M (Ctrl+Shift+M) to mute the microphone without ending the session.
Microsoft lists three conditions: Voice Mode is rolling out gradually, it requires an eligible individual GitHub Copilot plan, and it is not available with Copilot Business or Copilot Enterprise. Organizations can also switch off Copilot preview features by policy. The privacy section of the same page describes dictation; it does not say where Voice Mode audio is processed, so do not assume the on-device statement covers spoken conversations too.
5. JetBrains IDEs, Visual Studio and Copilot CLI: what to use instead
On Microsoft’s copilot-intellij-feedback tracker, three requests for a microphone in Copilot Chat for JetBrains IDEs were still open on October 3, 2026: issue 447 from July 2025, issue 1623 from April 2026 and issue 2024 from September 2026, which asks for parity with VS Code. JetBrains’ own AI Assistant and Junie have the same gap; YouTrack issue JUNIE-796 states there is currently no speech-to-text in either and lists a third-party speech-to-text tool as the workaround. So in IntelliJ IDEA, PyCharm, WebStorm or Rider, speech has to come from outside the IDE.
On Windows that means Win+H voice typing, which Microsoft says uses online speech recognition powered by Azure Speech and needs an internet connection and the cursor in a text box, or a desktop dictation app. On a Mac it means macOS Dictation or an app; the JetBrains-specific details are in our Mac guide. Visual Studio is in the same position: its Copilot Chat article covers the chat window, inline chat, images and Mermaid diagrams, with no voice control.
Copilot CLI is the exception. GitHub’s documentation says /voice transcribes entirely on your machine without sending audio over the network, runs on Windows (x64 and ARM64), Linux (x64) and Apple silicon Macs but not Intel Macs, and now offers a default multilingual model plus dedicated English and Spanish models. Hold Space for a short prompt or press Ctrl+X then V to toggle recording. Our Copilot CLI guide covers the setup step by step.
6. One dictation key for every Copilot host, recognized on-device
Where the host already has dictation, such as VS Code on an Apple silicon Mac or Copilot CLI, use it. IraVoice is for the gaps and for people who switch editors during the day. IraVoice 1.2.2 runs on macOS 13+ on Apple silicon, Windows 10 or 11 (64-bit), and Linux (beta), recognizes speech with NVIDIA Parakeet after a one-time model download, and types at the cursor of whatever has focus: Copilot Chat in a JetBrains IDE, the chat window in Visual Studio, VS Code, or a terminal running Copilot CLI. On a Mac and on Windows it pastes the text and then restores your clipboard.
On a Mac and on Windows it also spells the code names of the Git project the focused app is working in, so “get user by ID” arrives as getUserById and “pipeline dot swift” as Pipeline.swift. That works in VS Code and VS Code Insiders on both systems, in JetBrains IDEs on a Mac, and in IntelliJ IDEA, PyCharm, WebStorm, Rider and Visual Studio on Windows. Only names are kept, never source text; secrets and dependency folders are skipped, and nothing leaves the computer. On Linux, IraVoice is a beta that dictates into your editor without code names.
IraVoice has no Copilot integration and does not press Enter by default, so you read the prompt and send it yourself; what Copilot then does with it follows GitHub’s normal account and data terms. Hold Right Command on a Mac or Right Ctrl on Windows (Ctrl+Space on Linux), speak, and release. It costs $79 once for one computer, $109 for two, or $149 for three, with a 3-day free trial that needs no email or card.